Content uploaded by Tomasz Wojciechowski
Author content
All content in this area was uploaded by Tomasz Wojciechowski
Content may be subject to copyright.
M. Czechlowski, T. Wojciechowski „Journal of Research and Applications in Agric ultural Engineering” 2013, Vol. 58(1)
31
Mirosław CZECHLOWSKI, Tomasz WOJCIECHOWSKI
Uniwersytet Przyrodniczy w Poznaniu, Instytut Inżynierii Biosystemów
ul. Wojska Polskiego 28, 60-637 Poznań, Poland
e-mail: mczech@up.poznan.pl
THE UTILIZATION OF INFORMATION ABOUT LOCAL VARIABLE ENVIRONMENTAL
CONDITIONS TO PREDICT THE QUALITY OF WHEAT GRAIN DURING THE HARVEST
Summary
The study presents the correlation between the quality of winter wheat grain, understood as the content of protein, and
the parameters of harvested grain (moisture and yield) and locally variable environmental conditions (relative altitude, the
content of total Kjeldahl nitrogen, exchangeable phosphorus and potassium, magnesium, the pH coefficient, the content of
organic matter in soil). The results obtained on the basis of the data collected from 5 production fields of the total area of
112 ha give grounds for the conclusion that by application of multiple regression it is possible to construct a relatively pre-
cise model for prediction of the content of protein in wheat grain even on the basis of the measurement of easily obtainable
information about the relative altitude and yield. However, the effectiveness of the model will be limited to a small field ar-
ea. The construction of universal model using information about locally variable environmental conditions is difficult due to
the strong variability of the correlation between the analysed traits describing environmental conditions and the content of
protein in wheat grain.
The study was a part of the development project No. R12 0073 06 entitled "Development and validation of the technology
for separation grain stream during cereals selective harvesting ", financed by the Polish Ministry of Science and Higher
Education.
Key words: grain quality, multiple regression, environmental conditions, selective grain harvest, VIS-NIR spectroscopy
WYKORZYSTANIE INFORMACJI O LOKALNIE ZMIENNYCH WARUNKACH
ŚRODOWISKOWYCH W CELU PRZEWIDYWANIA JAKOŚCI ZIARNA PSZENICY
PODCZAS ZBIORU
Streszczenie
W pracy przedstawiono zależności pomiędzy jakością ziarna pszenicy ozimej, rozumianą jako zawartość białka, a parame-
trami zbieranego ziarna (wilgotność i wielkości plonu) oraz lokalnie zmiennymi warunkami środowiskowymi (względna wy-
sokość n.p.m., zawartość azotu ogólnego, wymiennego fosforu i potasu, oraz magnezu, współczynnik pH, zawartość materii
organicznej w glebie). Wyniki uzyskane w oparciu o dane zgromadzone na 5 produkcyjnych polach o łącznej powierzchni
112,78 ha pozwalają stwierdzić, że stosując regresję wieloraką można zbudować stosunkowo dokładny model do predykcji
zawartości białka w ziarnie pszenicy nawet w oparciu o pomiar łatwych do pozyskania informacji o względnej wysokości
n.p.m. i wielkości polonu, jednak jego skuteczność będzie ograniczona do niewielkiego obszaru powierzchni. Budowa uni-
wersalnego modelu wykorzystującego informacje o lokalnie zmiennych warunkach środowiskowych jest utrudniona ze
względu na silną zmienność zależności pomiędzy analizowanymi cechami opisującymi warunki środowiskowe, a zawarto-
ścią białka w ziarnie pszenicy.
Pracę zrealizowano w ramach projektu rozwojowego nr R12 0073 06 pt: „Opracowanie i walidacja technologii rozdziału
strumienia ziarna podczas selektywnego zbioru zbóż” finansowanego przez MNiSW
Słowa kluczowe: jakość ziarna, regresja wieloraka, warunki środowiskowe, selektywny zbiór zbóż, spektroskopia VIS-NIR
1. Introduction
The process of combine harvesting of crops is a basic
and widely applied method of harvesting cereals in large-
area farmlands [2]. During harvest all threshed grain, re-
gardless of its quality, comes into one grain container.
However, as is widely known, the farmland area may be
considerably diversified in terms of soil abundance in nutri-
ents or moisture [3, 4, 5, 10, 14] and it may be characterised
by individual landform, which influences the quality and
quantity of crops. Usually the highest content of protein can
be observed in terrains located at a high level, whereas the
yield is higher in terrains located at lower levels [3, 9].
So far research projects have usually been limited to
monitoring and recording the content of protein in harvest-
ed grain [7, 8, 15]. Until recently attempts to divide the
stream of grain were only made in stationary conditions in
granary [16]. However, in this solution the considerable
distance between the place of measurement and the place of
grain harvesting does not guarantee effective division due
to the multiple mixing of grain during harvest, reloading
and transport. As a result, the quality variance observed in
the field is lost and the grain represents averaged traits de-
scribing its quality. Therefore attempts to divide grain dur-
ing combine harvesting are justified [12, 13].
However, the authors of this study are of the opinion
that the decision algorithm used for controlling the process
of grain stream division, which is based only on the data
obtained from the spectrometer assessing the quality of
harvested grain, may be unreliable. As Maertens [8] proved,
this fact may be particularly evident in the case of very dy-
namic variations in the parameters describing grain quality
and simultaneous considerable delays of the signal due to
the time of the flow of grain through the threshing and
cleaning mechanisms of the combine harvester. At the same
time the authors think that the probability of making the
M. Czechlowski, T. Wojciechowski „Journal of Research and Applications in Agric ultural Engineering” 2013, Vol. 58(1)
32
right decision to send grain into one of the two chambers of
the grain tank in the harvester may be increased by using
the information about variable environmental conditions in
the direct neighbourhood of the harvester at work.
Therefore, due to the fact that the authors had a database
on grain parameter variability (moisture and winter wheat
grain yield) and the variability of soil parameters enhancing
yield (the content of total nitrogen, exchangeable phospho-
rus and potassium, magnesium, pH coefficient, the content
of organic matter in soil) they made an attempt to check
how much the parameters influence the content of protein
in winter wheat grain. However, the main goal of the study
was to select from the aforementioned data those which en-
able prediction of changes in the distribution of the content
of protein in wheat grain, on the one hand, and which will
be relatively easy to obtain during the work of a combine
harvester, on the other hand.
2. Material and methods
The research used the data obtained in 2011 from five
winter wheat production fields. The fields belong to three
experimental farms of Poznań University of Life Sciences
and they are situated in three different locations in the
western part of the Wielkopolska region (Poland). Soil and
grain samples for further analyses, grain spectrums and
basic data about the location of spectrums and samples
were obtained from the fields of the total area of 112 ha.
Soil samples were collected from the 0-0.25 m layer
eight weeks before harvesting. Mixed samples of the weight
of about 1000 g were collected in a regular square network,
where the area of one cluster was 1 ha for the fields larger
than 20 ha and 0.5 ha for the fields smaller than 20 ha. An
individual mixed sample was made up of 16-18 primary
samples. A precise grid of sample collection was made by
means of a GNSS Novatel Smart V1 receiver with a TDS
Recon recorder and 3R Area Pro field mapping software.
The laboratory analysis of the soil samples was made at the
Laboratory of National Chemical-Agricultural Station in
Poznań, accredited by the State Accreditation Centre for the
measurements which are the research subject. For the soil
samples the content of total nitrogen was labelled with the
Kjeldahl method, the content of absorbed phosphorus
(P2O5) and potassium (K2O) with the Egner-Riehm method,
magnesium (MgO) – with the Schachtschabel method, the
organic matter – with the Tiurin method and the pH – with
the potentiometric method in 1nKCl.
In order to obtain the spectrums of wheat grain and to
collect grain samples during harvest a ClaasLexion 480
combine harvester was used. It was equipped with an
AgroSpec spectrometer (tec5), a GNSS NovAtel PROPAK
V3 RT2 receiver with the RTK correction (ASG-EUPOS
system), a standard Claas system for yield measurement –
Quantimeter and an automatic grain sample collection sys-
tem constructed at the Institute of Biosystems Engineering
in Poznań. The absorption spectrums of radiation were rec-
orded by means of diffuse reflection at the wavelength
ranging from 400 to 2170 nm, with interpolated resolution
up to 2 nm. A contact measurement probe installed in the
measurement channel accumulating the grain sample col-
lected from the grain conveyor of the combine harvester
was used for this purpose. While the combine harvester was
working, 21.2 thousand spectrums were recorded and 599
grain samples were collected, for which the geographical
position and altitude AMSL were also recorded with sub
centimetre accuracy. The obtained values of altitude AMSL
were converted to relative altitude, where the lowest point
in the field was assigned the value of zero.
A Foss Infratec 1241 grain analyzer was used to meas-
ure the content of protein in the dry weight of the grain. The
grain moisture was measured according to PN-EN ISO in
all of the collected samples. On the basis of the results cali-
bration models were built by means of PLS method [1, 6,
11] (R2=0.75; RMSECV=0.59 for protein content in dry
mass and R2=0.85; RMSECV=0.58 for grain moisture),
which was available from Unscrambler X software (CAMO
Software AS). The prediction of the content of protein and
grain moisture was made in the post-processing mode on
the basis of the collected spectrums.
a)
b)
Fig. 1. Examples of interpolated maps showing the varia-
bility of: a) protein content distribution, b) altitude AMSL,
for the field No. P.50 at the Experimental Agricultural
Farm Przybroda
The data which were obtained with the aforementioned
methods and the data about the yield (dry), obtained from
the board computer of the combine harvester, were initially
prepared by means of a spread sheet (MS Office Excel).
Then, on their basis digital maps (Fig. 1) of spatial variabil-
ity of the aforementioned parameters were made with SMS
Advance Demo software (AgLeader). Next, the maps were
interpolated into a network sized 7 x 7 m and the data in
this form were exported to Statistica 10 package, where the
essential part of data analysis took place.
Multiple regression was applied to analyse of the influ-
ence of selected traits of locally variable environmental
conditions on the quality of winter wheat grain. Its aim was
to prove the significance α = 0.05 of the traits influencing
the quality of harvested grain, where the content of protein
was determined as quality (independent trait). The Pearson
M. Czechlowski, T. Wojciechowski „Journal of Research and Applications in Agric ultural Engineering” 2013, Vol. 58(1)
33
linear correlation analysis was also used to determine the
influence of individual independent traits on the dependent
trait. The aforementioned analyses were made independent-
ly for each of the five fields under investigation and for the
entire set of data.
3. Results
Although for the entire set of data the multiple regres-
sion analysis confirmed the statistical significance of α =
0.05 for all the independent traits in relation to the depend-
ent trait (the content of protein), for individual fields, which
were analysed separately, the presence of traits without sig-
nificance to the grain quality represented as the content of
protein was found. The following values were usually indi-
cated as statistically insignificant: relative altitude (B.21,
P.50), potassium content (P.2, P.50) and yield (B.21, P.2).
The comparison of the results of multiple regressions
(Table 1) shows that the adjustment of the regression model
to the set of data deteriorates along with the increase in the
area under investigation. This fact may indicate considera-
ble diversification of the character of the influence of indi-
vidual traits on the grain quality in different areas of the
fields. This observation is somewhat surprising in view of
the fact that the fields under investigation had relatively un-
diversified landform (usually slightly sloping in one direc-
tion) and small elevations (B.2 Δh=8.14m; B.21 Δh=7.42
m; P.2 Δh=3.78 m; P.50 Δh= 4.45 m; S.69 Δh=2.70 m).
The values of Pearson correlation coefficients in Table 2
also indicate considerable diversification of the strength of
the influence of individual traits on the quality of grain be-
tween individual fields. However, the obtained results point
to the fact that the content of protein in grain is most influ-
enced by the content of total nitrogen and organic matter in
soil. Simultaneously, the analysis of correlations between
individual traits for the entire set of data indicates a rela-
tively strong relationship between the traits (r = 0.80). At
the same time the content of total nitrogen and organic mat-
ter in soil are definitely negatively correlated with the rela-
tive altitude (r = 0.48 and r = 0.60 respectively). This rela-
tionship results in the observed negative correlation be-
tween the relative altitude and the content of protein.
The small influence of such traits as the content of
phosphorus and magnesium in soil on the content of protein
in wheat grain is confirmed by low values of partial and
semi-partial correlations (about 0.05) obtained for the entire
set of data. Similar low values of partial and semi-partial
correlations obtained for the dry yield are contrasted by the
high toleration value (1-R2=0.92) obtained for this variable.
It suggests the fact that in spite of the observed low correla-
tion between the content of protein and yield the trait con-
tributes some variability to the regression model.
On the basis of the aforementioned results the authors
made a decision to determine the variability of the accuracy
of the multiple regression model with a smaller number of
independent variables and a change in the cardinality of the
model set. On the basis of proved correlations between the
variables three sets of data were selected. The first con-
tained data on the content of total nitrogen and organic mat-
ter in soil, which was supplemented with the values of rela-
tive altitude and yield. The content of total nitrogen in soil
was excluded from the second set. The third set contained
only the data about the relative altitude and yield. Table 3
shows the obtained results.
4. Conclusions
The obtained results give grounds for the conclusion
that in the case of small sets of data (approximate field area
- 5 ha) it is also possible to obtain satisfying results of pre-
diction of protein content if the input data of the regression
model are limited to relative altitude and yield. This obser-
vation is valuable due to the fact that this information is
Table 1. A comparison of the results of multiple regressions for individual fields and for the entire set of data
Farm Field num-
ber
Field area
(ha)
Number of
records
Multiple
R
Multiple
R2
Corrected
R2
Standard error of
estimate
RGD Brody B.2 37.54 1270 0.78 0.61 0.61 0.73
RGD Brody B.21 53.69 9942 0.68 0.46 0.46 0.77
RGD Przybroda P.2 4.44 6715 0.42 0.17 0.17 0.38
RGD Przybroda P.50 4.98 1034 0.72 0.52 0.51 0.18
RGD Swadzim S.69 12.13 2256 0.67 0.45 0.45 0.78
EDS* 112.78 21218 0.45 0.20 0.20 0.98
* the entire data set
Table 2. A comparison of Pearson correlation coefficients between the content of protein and individual independent traits
Field
number
Yield mass
dry
Grain
moisture
Relative
altitude
Soil
N
t
otal
Soil
P2O5
Soil
K2O
Soil
MgO
Soil
OM
Soil
pHKCl
B.2 0.12 0.15 -0.36 0.00 -0.51 0.29 0.51 0.03 -0.59
B.21 0.00 0.52 -0.31 0.44 0.34 0.35 0.35 0.42 0.11
P.2 -0.03 -0.16 0.31 -0.18 -0.09 -0.14 -0.25 -0.22 0.15
P.50 0.13 -0.28 -0.29 0.19 0.12 0.23 -0.23 -0.43 0.36
S.69 0.01 -0.38 -0.14 0.33 -0.06 0.19 -0.01 0.37 0.22
EDS* -0.03 0.14 -0.18 0.31 -0.07 0.12 0.10 0.34 0.00
* the entire data set
M. Czechlowski, T. Wojciechowski „Journal of Research and Applications in Agric ultural Engineering” 2013, Vol. 58(1)
34
Table 3. A comparison of the results of multiple regressions for different data sets of independent variables and cardinality
of model sets
Field number Applied varia-
bles
Number of rec-
ords
Multiple
R
Multiple
R2
Corrected
R2
Standard error
of estimate
P.50 A N OM Y* 1035 0.68 0.47 0.46 0.19
A OM Y** 0.67 0.45 0.45 0.19
A Y*** 0.63 0.40 0.40 0.20
B.21 A N OM Y 9942 0.46 0.21 0.21 0.93
A OM Y 0.42 0.18 0.18 0.95
A Y 0.31 0.09 0.09 0.99
EDS* A N OM Y 21218 0.34 0.12 0.12 1.03
A OM Y 0.34 0.12 0.11 1.03
A Y 0.18 0.03 0.03 1.08
*ANOMY – relative altitude, total nitrogen, organic matter, yield, **AOMY – relative altitude, organic matter, yield,
***AY – relative altitude, yield
available in most of currently manufactured high-efficiency
combine harvesters, so it will not be necessary to bear high
costs to equip the harvester with additional sensors to ob-
tain the information.
As the field area increases, the accuracy of the regres-
sion model decreases rapidly. It can be at least partially
prevented by entering information about the variability of
the content of total nitrogen and organic matter in soil into
the model. Our calculations confirm the fact that the high
correlation between the data gives a possibility to omit one
of them without a significant loss to the accuracy of the re-
gression model. The authors think that it would be much
easier to equip a combine harvester with a sensor measuring
the content of organic matter in soil. It could use the NIRS
technology (Near Infrared Spectroscopy).
However, in the case of very large fields even the appli-
cation of this solution does not always guarantee appropri-
ately accurate prediction of the protein content so that it can
be used as supplementary information in controlling the
process of separation of the grain stream. For example, it is
noticeable at the moments when too dynamic variations of
the values returned by the spectrometer used for evaluation
of the grain quality make it difficult to determine the trend
of variations in the content of protein.
Therefore, further research will check the impact of lo-
cally variable soil moisture on grain quality, because the
lack of water, even with sufficient availability of minerals
compounds, may also result in a decrease in wheat grain
protein content.
5. References
[1] Dayal B. S., MacGregor J. F. Improved PLS Algorithms, J.
Chemom., 1997, 11, 73-85.
[2] Dreszer, K. A., Gieroba, J. & Roszkowski, A. Kombajnowy
zbiór zbóż. Warszawa, IBMER, 1998, (Chapter 2).
[3] Fiez, T.E., Miller, B.C. and Pan, W.L. Winter wheat yield
and grain protein across varied landscape positions. Agrono-
my Journal, 1994, 86: 1026-1032.
[4] Jadczyszyn, T. Zmienność gleb, plonów i potrzeb na-
wozowych w granicach pola produkcyjnego. Inżynieria
Rolnicza, 2001, 13 (33), 168-173.
[5] Kollárová, K., Krajčo, J., Plačko M. & Rutkowski K. Ocena
zmienności przestrzennej wilgotności gleby na podstawie
map konduktywności elektrycznej. Inżynieria Rolnicza,
2007, 6(94), 73-80.
[6] Lindgren F., Geladi P., Wold S. The kernel algorithm for
PLS, J. Chemom., 1993, 7, 45-59.
[7] Long, D. S., Engel, R. E., & Carpenter, F. M. On-Combine
Sensing and Mapping of Wheat Protein Concentration, Crop
Management (On-line), 2005, Published 27 May 2005.
[8] Maertens, K., Reyns, P., De Baerdemaeker, J. On-line meas-
urement of grain quality with NIR technology. Transactions
of the ASAE, 2004, 47(4), 1135-1140.
[9] Moore, I.D., Gessler, P.E., Nielsen, G.A. and Peterson, G.A.
Soil attribute prediction using terrain analysis. Soil Science
of America Journal, 1993, 57:443-452.
[10] Nolan, S.C., Goddard, T.W., Penney, D.C. & Green, F.M.
Yield response to nitrogen within landscape classes. Proceed-
ings of the 4th International Conference on Precision Agri-
culture, St. Paul, MN, July 19-22, 1998, 479-485.
[11] Rannar S., Lindgren F., Geladi P., Wold S. A PLS kernel al-
gorithm for data sets with many variables and fewer objects,
Part 1: Theory and Algorithm, J. Chemom., 1994, 8, 111-
125.
[12] Risius, H., Hahn, J., Huth, M., Korte, H., Luetke Harmann,
T. Near Infrared Spectroscopy for Sorting Grain according to
Specified Quality Parameters on a Combine Harvester. In B.:
67th International Conference on Agricultural Engineering
LAND.TECHNIK AgEng (pp.187-192), Stuttgart-
Hohenheim: VDI, Germany, 2008,
[13] Risius, H., Hahn, J., Korte, H., Monitoring of grain quality
and segregation of grain according to protein concentration
threshold on an operating combine harvester. In B.: Book of
Abstracts XVII.th World Congress of the International
Commission of Agricultural and Biosystems Engineering
(CIGR | SCGAB) (pp. 28), Québec City: QC, Canada. 2010.
[14] Stewart. C.M., Mc Bratney, A.B. &Skerritt, J.H. Site-specific
Durum wheat quality and its relationship to soil properties in
a single field in Northern New South Wales. Precision Agri-
culture, 2002, 3, 155-168.
[15] Taylor, J.; Whelan, B.; Thylén, L.; Gilbertsson, M.; Hassall,
J. Monitoring wheat protein content on-harvester - Australian
experiences. In B. J. V. Stafford (Eds.) Precision agriculture
'05, 5th European Conference on Precision Agriculture, Con-
ference paper (pp. 369-375), Uppsala, Sweden, 2005.
[16] Thylen, L., Gilbbertsson, M., Rosenthal, T., Wrenn, S. Sort-
ing of Grain on the Farm - Experiences with an Online Pro-
tein Sensor. In B: D.E. Maier (Eds.) International Quality
Grains Conference (pp. 1-8), Indianapolis: Purdue Universi-
ty, USA, 2004.
The paper was published in the Papers Book of International Con-
ference of Agricultural Engineering. CIGR-AgEng2012. July 8-
12, 2012.Valencia. Spain.
