Available online at www.sciencedirect.

com

ScienceDirect
Energy Procedia 47 (2014) 156 – 165

Conference and Exhibition Indonesia Renewable Energy & Energy Conservation

[Indonesia EBTKE CONEX 2013]

Fresh Tuber Yield Stability Analysis of Fifteen Cassava Genotypes

Across Five Environments in East Java (Indonesia)
Using GGE Biplot

Kartika Noerwijatia,*, Nasrullahb, Taryonob, and Djoko Prajitnob

a
Indonesian Legumes and Tuber Crops Research Institute, Kendalpayak km 8, PO BOX 66, Malang 65101, East Java, Indonesia
b
Gadjah Mada University,Jl. Bulaksumur, Yogyakarta 55281, Indonesia

Abstract

The research aimed is to determine the yield stability of 15 cassava genotypes using GGE biplot. The study was conducted at five
locations starting in November 2010 to August 2011.The results showed that environment, genotype, and genotype-by-
environment interactions significantly affect the yield. Environment gives the most effect (64.69%), followed by genotype-by-
environment interaction effect (6.53%), and genotype effect (4.94%). CMM 03038-7 has the highest yield among other clones,
and higher than the controls clones (UJ5, Malang 6, and Adira 4), but not significantly different from Malang 4. GGE biplot
identified that CMM 03038-7 is the most stable clones with high yield.

© 2014
© 2014The
TheAuthors.
Authors. Published
Published by by Elsevier
Elsevier Ltd.Ltd.
Selectionand
Selection andpeer-review
peer-review under
under responsibility
responsibility ofScientific
of the the Scientific Committee
Committee of Indonesia
of Indonesia EBTKEEBTKE Conex 2013.
Conex 2013

Keywords: multi-environment trial; genotype-by-environment interaction; stability analysis; GGE biplot

Nomenclature

GGE Genotype + Genotype-by Environment

CMM Cross Manihot Malang
GEI Genotype-by-Environment Interaction
PC Principle Component
SVD Singular Value Decomposition
AMMI Additive Main Effect dan Multiplicative Interaction

* Corresponding author. Kartika Noerwijati Tel.: +62-34-180-1468; fax: +62-34-180-1496.

E-mail address: tika_iletri@yahoo.com

1876-6102 © 2014 The Authors. Published by Elsevier Ltd.

Selection and peer-review under responsibility of the Scientific Committee of Indonesia EBTKE Conex 2013
doi:10.1016/j.egypro.2014.01.209
 Kartika Noerwijati et al. / Energy Procedia 47 (2014) 156 – 165 157

SREG Site Regression

G Genotype
S Site
AEC Average Environment Coordinate
AEA Average Environment Axis
AOE Average Ordinate Environment
MET Multi Environment Trial

1. Introduction

Approximately 70% of cassava production in Indonesia is edible to food. Cassava demand has increased with the
development of industries that use cassava as raw materials including bioethanol industry. The efforts to increase
cassava production must be done so that the need of cassava as a food was not bothered where cassava is a priority
commodity in the food diversification program to support food security. Cassava demand has increased 5% to 7%
per year. This can be achieved through increased productivity of 3% to 5% per year and the expansion of planting
area approximately 10% to 15% per year [1].
Cassava tuber yield is strongly influenced by environmental conditions, varieties, and harvesting time. The
magnitude of genotype-by-environment interaction (GEI) is the cause of the cassava clones should be planted in a
suitable environment so that the potential yield of a genotype to be maximum. Genotype testing in various
environments will help to identify cultivar with broad adaptation and specific adaptations [2].
There are three types of GEI, i.e (1) no interaction, (2) no crossover intercation, and (3) crossover interaction
[3]. Crossover interactions are the most important for breeders as they imply that the choice of the best genotype is
determined by the environment [4]. Only crossover interaction affects genotype selection. To determine crossover
interaction, both genotype (G) and GEI must be considered [3].
Multilocation trials is very important to be done with the purpose of which are (1) to compare the genotypes
performance in multi environments and in the specific environment, (2) to estimate GEI component to measure its
impact on the heritability and selection, (3) to select test sites and determines the environment in a broader scope,
(4) to identify genotypes with specific adaptations, and (5) to determine breeding objectives [3].
GGE biplot technique is an analytical technique of GEI involving the effect of genotype (G) and the effect of
genotype-by-environment (GE) [3]. The biplot is constructed by the first two symmetrically scaled principal
components (PC1 and PC2) derived from singular value decomposition (SVD) of multi-environment trial data.
Biplots can be multidimensional, but two-dimensional biplots, using only the first and the second PCs, are most
common [5].
The analyses of GEI data include three major aspects: (1) mega-environment analysis, (2) test-environment
analysis, and (3) genotype evaluation [3]. The “which-won-where” view of GGE biplot is an effective visual tool in
mega-environment analysis [5]. GGE biplot technique (developed after AMMI biplot technique) has several
advantages when compared with AMMI techniques include (1) the graph of the GGE biplot better than AMMI1
graph in mega-environment analysis and genotype evaluation because it explains more G + GE, and has the inner
product of property of biplot, (2) the discriminating power vs representativeness view of the GGE biplot is effective
in evaluations test environments, which is not possible in AMMI analysis [6].
GGE biplot can be used to analysis of the mega-environment [7,8,9], genotype evaluation [10], test-
environment evaluation [7,11], and analysis of heterotic patterns [12]. These aspects cause the GGE biplot become
very popular as the most comprehensive tool in quantitative genetics and plant breeding.
The commonly used GGE biplot is based on the Sites Regression (SREG) linear-bilinear (multiplicative) model
[8,13]. In the SREG model, the main effect of genotype (G) plus GE interaction were absorbed into the bilinear
terms [14], which are the two sources of variation that are relevant to cultivar evaluation [15]. These factors are
graphically shown through a GGE biplot, which is used in the visual evaluation of both genotypes and environments
[16].
The purpose of this research is to determine the stability of fresh tuber yield of 15 cassava genotypes grown in
five locations using GGE biplot technique.
158 Kartika Noerwijati et al. / Energy Procedia 47 (2014) 156 – 165

2. Material and method

The research was conducted in November of 2010 to August of 2011 at five locations : Kediri, Ponorogo,
Probolinggo, Malang, and Mojokerto (East Java, Indonesia). The research was conducted using a randomized block
design with three replications. The treatments are 15 cassava genotypes consisting of eleven cassava clones namely
CMM 03025-43, CMM03036-7, CMM 03036-5, CMM 03038-7, CMM 03094-12, CMM 03094-4, CMM 03095-5,
CMM 02040-1, CMM 02033-1, CMM 02035-3, CMM 02048-6, and four control varieties, namely UJ5, Malang 6,
Malang 4, and Adira 4.
Cassava is planted in a plot size of 5 m × 5 m with a spacing of 100 cm × 80 cm. Cassava cuttings about 20 cm
long are planted with the vertically position of cuttings. Fertilization was given twice, at one month after planting
with a dose of 100 kgs ha-1 Urea + 100 kgs ha-1 SP36 + 100 kgs ha-1 KCl, and at three months after planting with
100 kgsha-1 Urea. Weeding was performed twice, at one and three months after planting. The activities to improve
the ridge were carried out before fertilization. Removal shoots with leaves two best buds performed at two months
after planting. Harvesting was done when the plant was 10 months. Observations were made on fresh tuber yield per
plot. Tuber yield per plot converted to fresh tuber yield per hectare.
Combined analysis of variance was performed on the fresh tuber yield data. The equation for the combined
analysis of variance is

Y ij P  e j  gi  ( ge)ij  H ij (1)

where Y ij is the yield mean of the ith genotype at the jth environment (i = 1,2,...,v dan j = 1,2,...,s), v is the number of
genotypes, s is the number of environments, μ is the grand mean, gi is the genotype effect, ej is the environment
effect, (ge)ij is the genotype-by-environment interaction effect, and εij is the error associated to the model.
If there was a significantly different in genotype-by-environment interaction was followed by GGE biplot
analysis to assess the stability of tuber yield of 15 cassava genotypes that tested. The equation for GGE biplot is
t
Y ij P  e j  ¦ Ok D ik J jk  H ij (2)
k 1
where Y ij is the yield mean of the ith genotype at the jth environment (i = 1,2,...,v dan j = 1,2,...,s), v is the number of
genotypes, s is the number of environments, μ is the grand mean, ej,is the environment effect, O k is the eigenvalue of
the PCA analysis axis k, D ik ( D ik ,..., D gk ) and J jk ( J jk ,…, J ek ) are the genotype and environment principle component
scores for axis k, t is the number of principle component retained in the model, and H ij is the error.

3. Result and discusion

The results of the combined analysis of variance showed that environment, genotype and genotype-by-
environment interactions significantly affect fresh tuber yield. Environment was the most important source for tuber
yield variation (64.69%), followed by genotype-by-environment interaction effect (6.53%), and the genotype effect
(4.94%) (Table 1). It showed that cassava tuber yield is strongly influenced by environmental factors. However, the
large environment effect is not relevant to cultivar evaluation [3]. Only genotype (G) and interaction genotype-by-
environment (GE) are relevant to cultivar evaluation. Therefore, for cultivar evaluation, it is essential to remove
environment effect (E) from data and focus on G and GE. Further explained that another important point in cultivar
evaluation is that G and GE effect must be considered simultaneously to make any meaningful selection decisions.
GGE biplot analysis divides the sum of squares of genotype plus genotype-by-environment interactions into
several principle components that are PC1, PC2, and PC3, each of which has a proportion of 55.98%, 24.44%, and
10.27% of the sum of squares genotype and interaction. However, only the first two components (PC1 and PC2),
which have a significant effect (Table 2). PC1 and PC2 values both the genotype and the environment are presented
in Table3.
 Kartika Noerwijati et al. / Energy Procedia 47 (2014) 156 – 165 159

Table 1. Combined anova for fresh tuber yield of fifteen cassava genotypes tested
at five environments.
Source of variance Df Sum of Means F value Proportion
Square Square (%)
Env 4 49843.52 12460.88 263.46** 64.69+
Rep (Env) 10 2495.11 249.51 5.28**
Genotype 14 3806.41 271.89 5.75** 4.94+
Genotype-by- Env 56 5029.75 89.82 1.90** 6.53+
Plant Number 1 134.29 134.29 2.84
Error 134 6337.89 47.30
Total 219 77050.72
** Siginifantly different (p < 0,01)
+
Proportion to total sum of square

GGE biplot is presented with two principle components explaining a total of 80.42% GGE variation (PC1
55.98% and PC2 24.44%) (Table 2). The first principle component (PC1) is represented in X axis. The genotype that
have higher PC1 values are considered be more productive. The second principle component (PC2) is represented
on the Y axis and describes the genotype stability.

Table 2. GGE analysis for fresh tuber yield .

Principle Eigenvalues Porcentase (%) Porsenac (%) Probability
Component
PC1 5205.53 ** 55.9805 55.980 0.00000
PC2 2272.55 ** 24.4391 80.420 0.00016
PC3 954.89 10.2689 90.689 0.10670
PC4 734.72 7.9013 98.590 0.17428
Residual 131.13 1.4102 100.000 0.97117
Total 9298.82
** Significantly different (p < 0.01)

3.1. Performance of different genotypes at specific environment

Figure 1 shows the comparison of relative performance of all genotypes at S1 (Kediri). The first step is to make a
straight line passing through the biplot origin to S1 marker, to make the S1 axis. The second step is to make a line
was perpendicularly drawn from each genotype toward the S1 axis. The line which passed through the biplot origin
and was perpendicularly to S1 axis, separates genotypes that have yield higher than the mean from genotypes that
have yield lower than the mean [3, 8]. Three genotypes that have highest yield at S1 are Malang 4 (G3), CMM
03036-7 (G6), and CMM 03038-7 (G8), whereas genotypes that have low fresh tuber yield at S1 are UJ5, CMM
02048-6 (G15), CMM 03095-5, (G11), and CMM 03094-12 (G9). There are six genotypes that have fresh tuber
yield above the mean, and nine genotypes below the mean (Fig.1).
The distance from the biplot origin to the marker of an environment is called the environment’s vector. The
lenght of the vector is a measure of the environment’s ability to discriminate among genotype. The short vector,
relative to the biplot sizes, implies that all genotypes tend to have similar yield in the associated environment [3].

3.2. Relative adaptation of a specific genotype across environment

One of the activities in the genotype testing was to determine which environment is most suitable for a
genotype. GGE biplot can be used for this purpose. A line was drawn pass through the biplot origin and the marker
of G8 (G8 is used as an example), which may be called the G8 axis. The environment are ranked along to the G8
axis in the direction indicated by the arrow. Based on Fig.2, G8 performed well at S1, followed by S4, S3, S5 and
S2.
160 Kartika Noerwijati et al. / Energy Procedia 47 (2014) 156 – 165

A line that is perpendicularly to the G8 vector and pass through the biplot origin separates the environment in
which G8 yielded below the mean from those in which G8 yielded above the mean. At all environments, G8 always
had yield above the mean of each environment (Fig.2).

Table 3. The mean of fresh tuber yield (t ha-1), PC1 and PC2 values of fifteen
genotypes and five environments.
Genotypes The mean of fresh PC1 PC2
tuber yield(t ha-1)
UJ5 (G1) 25.22 -1.64 -0.68
Malang 6 (G2) 32.58 1.33 0.91
Malang 4 (G3) 37.79 3.37 -0.96
Adira 4 (G4) 31.51 0.28 2.35
CMM 03025-43 (G5) 26.91 -0.91 0.07
CMM 03036-7 (G6) 31.52 1.10 0.87
CMM 03036-5 (G7) 29.93 -0.82 2.59
CMM 03038-7 (G8) 37.52 2.69 0.55
CMM 03094-12 (G9) 24.40 -1.32 -1.64
CMM 03094-4 (G10) 34.55 1.81 -2.28
CMM 03095-5 (G11) 23.95 -2.24 -0.04
CMM 02040-1 (G12) 28.13 -0.92 -1.30
CMM 02033-1 (G13) 29.49 0.04 -0.86
CMM 02035-3 (G14) 24.16 -0.68 1.10
CMM 02048-6 (G15) 26.69 -2.10 -0.68
Kediri (S1) 54.84 4.65 1.76
Ponorogo (S2) 7.79 0.45 0.41
Probolinggo (S3) 31.00 2.25 2.70
Malang (S4) 37.08 3.63 -4.11
Mojokerto (S5) 18.28 1.28 0.37

Fig. 1. The performance of different genotypes at S1 (Kediri).

3.3. Comparison of two genotypes performance all environments

With the same principle, the performance of two genotypes can be compared on the GGE biplot. To compare two
genotypes, for example, G4 and G10, a line that connected the markers of G4 and G10 was drawn, and a line that
was perpendicularly to the first line that passes through the biplot origin was drawn. The perpendicular line divides
the environment into two groups, with each genotype yielding better than the other within its respective side of the
 Kartika Noerwijati et al. / Energy Procedia 47 (2014) 156 – 165 161

perpendicular line. Fig.3 showed that one environment, S3, are on the same side with genotype G4. Thus, G4 would
yield better than G10 at S3, while G10 yielded better at the other environment (namely S1, S2, S4 and S5).

Fig. 2. The relative adaptation of a specific genotype in different environment.

Fig. 3. The compariosn of two genotypes.

3.4. Identification of the best genotypes for each environment

The polygon of the GGE biplot provides an effective and elegant tool for visualizing the “which-won-where”
patern. The polygon not only shows the best genotype for each test environment but also devides the test
environment into several groups [3].Two criteria are required to suggest existence of different mega-environments.
First, there are different winning genotypes in different test environments. Second, the among-group variation
162 Kartika Noerwijati et al. / Energy Procedia 47 (2014) 156 – 165

should be significantly greater than the within-group variation, a common criteria for clustering [17]. Graphically,
different mega-environments should consist of test environments.
The polygon is drawn joining the genotype that are located farthest from the biplot origin (0,0) that served as
corners of polygon, so that all genotypes are located in the polygon. Then, perpendicular line to each side of the
polygon is drawn passing through the biplot origin, so the environment are divided into several sectors, each sector
with different corner genotype. Within a sector, genotype which is located at the top of polygon is the best
genotypes in all environments that located in the sector [5,3].
There are formed eight sectors with the corner genotypes are G8, G4, G7, G11, G15, G9, G10, and G3. Only
three sectors contained environment, thus formed three mega-environments (Fig.4). Mega-environment is the group
of environments that share the same best genotype(s) identified as being located at the corner of polygon [8]. Based
on mega-environment that formed can be seen that Adira 4 (G4) is high yielding genotype at Probolinggo (small
mega-environment). Malang 6 (G2) and CMM 03036-7 (G6) are high yielding genotypes at Kediri (S1) and
Ponorogo (S2). Malang 4 (G3), CMM 03038-7 (G8), and CMM 03094-4 (G10) are high yielding genotypes at
Malang (S4) and Mojokerto (S5). And vice versa, genotype and environment at two sectors that are opposite to the
farthest distance showed that genotype had the largest negative interactions in the environment. For example, CMM
02048-6 (G15) has a low tuber yield at Kediri (S1) and Probolinggo (S3). CMM 03095-5 (G11) has low tuber yield
in Ponorogo (S2) (Fig.4). A genotype planted outside its mega-environment frequently suffers yield reduction [17].
Deviding the target environment into different mega-environments and developing different genotypes in
different mega-environment is the best way to utilize genotype-by-environment interaction [3]. Mega-environment
is formed can be simple or complex [7].

Fig. 4. The mega-environment and their winning genotypes.

3.5. Ideal genotype identification

Estimation of yield and genotype stability was done using AEC (Average Environment Coordinate) [18], where
the average environment is defined by the average values of PC1 and PC2 of all environments and is presented with
a red circle on the biplot graph. A line passing through the origin of biplot and AEC point was made and is called
the average environment axis (AEA). The line which is perpendicular to the AEA line and pass through the origin is
called the average ordinate environment (AOE). This line separates genotypes that have fresh tuber yield higher than
grand mean (right of the line) with genotypes that have a fresh tuber yield lower than grand mean (left of the line).
Genotypes that have higher tuber yield than the grand mean are Malang 6 (G2), Malang 4 (G3), Adira 4 (G4), CMM
 Kartika Noerwijati et al. / Energy Procedia 47 (2014) 156 – 165 163

03036-7 (G6), CMM 03038-7 (G8), and CMM 03094-4 (G10). Genotypes that have lower tuber yield than the grand
mean are UJ5 (G1), CMM 03025-43 (G5), CMM 03036-5 (G7), CMM 03094-12 (G9), CMM 03095-5 (G11), CMM
02040-1 (G12), CMM 02033-1 (G13), CMM 02035-3 (G14), and CMM 02048-6 (G15) (Fig.5).
Stability of the genotypes depends on their distance from the AEA line. The genotypes closer to the AEA line
are more stable than others. The results of this study indicate that the most stable genotype and high tuber yield is
CMM 03038-7 (G8). Genotype CMM 03025-43 (G5) and CMM 03095-5 (G11) are also a stable genotype, but it
have low tuber yield. Other genotypes that relative stable and have high tuber yield are Malang 6 (G2) and CMM
03036-7 (G6).
Ideal genotypes are genotypes that should have large PC1 scores (high mean yield) and small (absolute) PC2
score (high stability) [5, 7, 3]. Projection of the ideal genotype on the AEA line is equal to the longest vector of all
genotypes and its projection on the AOE line is obviously zero [19]. Based on criteria of ideal genotype as already
mentioned above, the closest genotype to the ideal genotype is CMM 03038-7 (G8) (Fig.6).

AEC

Stable

Unstable

Low yield High yield

Fig. 5. GGE biplot base symmetrical scale with AEC.

3.6. Identification of ideal test environments

Ideal environments are environments that should have small (absolute) PC2 scores (more representative of the
overall environment) and large PC1 score (more power to discriminate genotype in terms of the genotypic main
effect) [5,7]. Ideal environment can be used to as a reference for genotype selection in the multi-environment trial
[20]. The projection of ideal environment on the AEA line is equal to the longest vector of all environment (the most
discriminating environment) and its projection on the AOE line axis was zero, meaning that is absolutely
representative of average environments [19]. The ideal environment that represented by the green small circle is the
most discriminating of genotypes and yet representativeness of the other test environments. Accordingly, S1
(Kediri) is the test environment that is closest to the ideal environment, so it can be categorized that Kediri is the
most ideal environment compare to other environments (Fig.7)
164 Kartika Noerwijati et al. / Energy Procedia 47 (2014) 156 – 165

AEC

Ideal genotype

Fig. 6. The comparison of all genotypes with ideal genotype.

AEC

Ideal environment

Fig. 7. The comparison of all environments with ideal environment.

4. Conclusion

According to GGE biplot analysis, CMM 03038-7 (G8) is the most stable clone across five environments with
high tuber yield, and Kediri (S1) is the ideal environment to discriminating genotypes.
As has been reported previously by many authors, GGE biplot can make biplot analysis of MET data very easy,
informative, and interesting.
 Kartika Noerwijati et al. / Energy Procedia 47 (2014) 156 – 165 165

Acknowledgements

The authors are grateful to Agency of Agriculture Research and Development of Indonesian Agriculture Ministry
for providing a doctoral scholarship fund.

References

[1] Wargiono J. Potential areas for cassava development as bioethanol raw materials. Bogor: Indonesian Center for Food Crops Research and
Development; 2007. Unpublished.
[2] Baafi E, Safo-Kantanka O. Agronomic evaluation of some local elite and released cassava varieties in the forest and traditional ecozones
of Ghana. Asian Journal of Agriculture Research 2008; 2(1):32-36.
[3] Yan W, Kang MS. GGE biplot analysis, a graphical tool for breeders, geneticists, and agronomists. Boca Raton, London, New York,
Washington D.C: CRC Press; 2003.
[4] Malosetti M, Ribaut JM, van Eeuwijk FA. The statistical analysis of multi-environment data: modeling genotype-by-environment
interaction and its genetic basis. Front Physiol 2013; 4:44.
[5] Yan W, Hunt LA, Sheng Q, Szalvincs Z. Cultivar evaluation and mega-environment investigation based on the GGE biplot. Crop Science
2000; 40:597-605.
[6] Yan W, Kang MS, Ma B, Woods S, Cornelius PL. GGE biplot vs AMMI analysis of genotype-by-environment data. Crop Science 2007;
47:643-655.
[7] Yan W, Racjan I. Biplot evaluation of test site and trait selection of soybean in Ontario. Crop Science 2002; 42:11-20.
[8] Samonte SOPB, Wilson LT, Chang AM, Medley JC. Targeting cultivars onto rice growing environment using AMMI and SREG GGE
biplot analysis. Crop Science 2005; 45:2414-2424.
[9] Yan W, Tinker NA. An integrated system of biplot analysis for displaying, interpreting, and exploring genotype-by-environment. Crop
Science 2005; 45:1004-1016.
[10] Malvar RA, Revilla P, Butron A, Gouesnard B, Boyat A, Soengas P, Alvarez A, Ordes A. Performance of crosses among French and
Spanish maize populations across environments. Crop Science 2005; Vol. 45. No. 3:1052-1057.
[11] Blanche SB, Myers GO. Identifying descriminating location for cultivar selection in Lousiana. Crop Science 2006;4:946-949.
[12] Yan W, Hunt LA. Biplot analysis of diallel data. Crop Science 2002; 42:21-30.
[13] Cornelius PL, Crossa J, Seyedsadr MS. Statistical test and estimates of multiplicative models for GE interaction. In: Kang, M.S, Gauch
HG, editors. Genotype-by-environment interaction. Boca Raton: CRC Press; 1996
[14] Crossa J, Cornelius PL, Yan W. Biplot of linear-bilinear models for studying genotype × environment interaction. Crop Science 2002;
42:619-633.
[15] Gauch Jr HG, Zobel RW. AMMI analysis of yield trials. In Kang MS, Gauch HG, editors. Genotype-by-environment interaction. Boca
Raton, FL: CRC Press; 1996. p.1-40.
[16] Yan W. Cultivar evaluation and mega-environment investigation based on the GGE-biplot. Crop Science 2000; 40:597-605.
[17] Gauch HG, Zobel RW.Identifying mega-environment and targeting genotypes. Crop Science 1997; 37:311-326.
[18] Yan W. GGE biplot – a windows application for graphical analysis of multi-environment trial data and other types of two way data.
Agronomy Journal 2001;93:1111-1118.
[19] Karimizadeh R, Mohammadi M, Sabaghn N, Mahmoodi AA, Roustami B, Seyyedi F, Akbari F. GGE biplot analysis of yield stability in
multi-environment trials of Lentil genotypes under rainfed condition. Notulae Scientia Biologiae 2013; 5(2):256-262.
[20] Farshadfar E, Zali H, Mohammadi R. Evaluation of phenotypic stability in Chickpea genotypes using GGE-biplot. Annal of Biological
Research 2011; 2(6):282-292.