Presented by Kassaye Negash and Kidane Tumsa (National Lowland Pulses Research Program at Melkassa Agricultural Research Center-EIAR) at the First Bio-Innovate Regional Scientific Conference, Addis Ababa, Ethiopia, 25-27 February 2013
Swan(sea) Song – personal research during my six years at Swansea ... and bey...
Line by environment interaction, yield stability and grouping of test locations for navy bean variety trial in Ethiopia
1. Line by environment interaction, yield
stability and grouping of test locations for
navy bean variety trial in Ethiopia
Kassaye Negash and Kidane Tumsa
National Lowland Pulses Research Program at Melkassa
Agricultural Research Center-EIAR
First Bio-Innovate Regional Scientific Conference
United Nations Conference Centre (UNCC-ECA)
Addis Ababa, Ethiopia, 25-27 February 2013
3. Importance of beans in Ethiopia
Beans are produced by about 2.5 millions households across Ethiopia
4. Increased bean production and productivity
Area Trend in bean production and area Production
Trend bean yield (t/Ha) (tons)
1.6 400,000 300,000
1.487 362,890
1.4 350,000
250,000
1.2 300,000 244,012
183,800
181,600 200,000
1 250,000
0.936
Production (t/Ha)
0.8 0.823
200,000 150,000
172,150
0.6 0.615 119,900
150,000
100,000
0.4 111,750
100,000
98,670
50,000
0.2
50,000
0
- -
2002/3 2003/4 2004/5 2009/10
2002/3 2003/4 2004/5 2009/10
Trend in bean production (Qt/Ha) Production (tons) Area
Sources : CSA reports
5. Trend of quantity and revenue of white pea beans exported to international markets
between 2005 - 2010
Quantity Quantity exported (t)
exported Trend in value of bean export
(tons)
(USD)
80000 Revenue
75864 74762
(USD)
68452 68638
70000
60834 $50,000,000 $49,046,107
$49,654,516
60000 $45,000,000 $44,747,590
$40,000,000
49679 $35,000,000
$36,229,556
50000
$30,000,000
$25,000,000
$20,220,954
40000 $20,000,000
$15,000,000
$8,146,125
30000 $10,000,000
$5,000,000
$0
20000
2005 2006
2007
2008
10000 2009
2010
0
2005 2006 2007 2008 2009 2010 Year
Sources : CSA reports Rev…
6. Some major impacts of the bean program
Private sector investment Government investment and
Farmer investment and and employment creation supportive policies
benefit Beans listed on ECX –
2004:
2ha
Better return to farmers : more than
2010: 30 600 % price increase between 2003
ha and 2011 : USD 120/ton to USD 800
7. Introduction ---
The major objective of breeding of beans is to
achieve higher and stable yield of the crop
Multi-environment trials are typically used in crop
improvement to evaluate materials across a range of
sites representing target environments for the crop
However, GEI change the relative performance of
genotypes across sites
8. Introduction ---
Understand the nature of GEI is important for testing
and selecting superior genotypes
Key concept in G x E analysis is genotype stability and
by definition, genotypes exhibiting a high degree of
GEI are unstable across sites and vice versa
In this study, AMMI statistical model was used to
study the nature of GEI among common bean lines
evaluated in nine locations during 2010 to 2011 main
crop growing season
9. Introduction ---
AMMI model is a recently preferred statistical model
to analyze multi-environment varietal trials effectively
and efficiently, where there is a usual occurrence of
GEI
AMMI is combining ANOVA for additive main effects
and uses PCA to partition the multiplicative structure
of the interaction
The ANOVA model partitions the total sum of squares
(SS) into the components: E, G and GEI without
further partitioning the interaction component making
interpretation difficult in terms of significance of
genotypes across different environments;
10. 2. Objectives
The objectives of this study were:,
1. to estimate the components of variance
associated with GE interaction and to
determine their effects
2. to compare the various statistics to
determine the most suitable method for
assessing navy bean line’s yield stability in
the major bean growing areas of Ethiopia
11. 3. Materials and Methods
Experiment was conducted in the main
growing seasons of 2010 to 2011
Locations were:
Melkassa, Alemtena, Areka, Haramaya, J
imma, Bako, Pawe, Sirinka and Assossa
The locations have diverse agro-
ecological characteristics as annual
rainfall, temperature and altitude
12. 3. Materials ---
Sixteen navy bean lines including released
two varieties (as checks) were used in this
study
RCBD with 3 reps was used at each location
Net size of the experimental unit/plot was
6.4 sqm
Data were collected on grain yield per plot
from which grain yield per hectare was
estimated at 14% moisture content
13. Table 2. Descriptive information on the name and codes of the 16 cowpea varieties
Line Code Number Line Name
1 ICA Bunsi x S x B 405/1C-C1-1C-1
2 ICA Bunsi x S x B 405/1C-C1-1C-3
3 ICA Bunsi x S x B 405/1C-C1-1C-13
4 ICA Bunsi x S x B 405/1C-C1-1C-14
5 ICA Bunsi x S x B 405/1C-C1-1C-23
6 ICA Bunsi x S x B 405/1C-C1-1C-30
7 ICA Bunsi x S x B 405/1C-C1-1C-37
8 ICA Bunsi x S x B 405/1C-C1-1C-51
9 ICA Bunsi x S x B 405/1C-C1-1C-58
10 ICA Bunsi x S x B 405/1C-C1-1C-69
11 ICA Bunsi x S x B 405/1C-C1-1C-70
12 ICA Bunsi x S x B 405/1C-C1-1C-80
13 ICA Bunsi x S x B 405/1C-C1-1C-87
14 ICA Bunsi x S x B 405/1C-C1-1C-88
15 Awash - 1
16 Awash melka
14. 3. Materials ---
Statistical analyses:
ANOVA was done for each location
separately
Data transformed to fix failures of
assumptions (normality and homogeneity of
error variances)
Combined ANOVA was done according to
the best AMMI model (by GenStat 14th
edition)
15. Statistical analyses
Mean yield data from each environment was
used for most of the stability analysis methods
(by AgrobaseTM 1999 software package)
The effect of GEI on the yield is then
determined by AMMI analyses (Gauch, 1993;
2007)
16. Statistical analyses ---
AMMI first fits additive effects for G and E by the
usual additive analysis of variance procedure, and then
fits multiplicative effects for GEI by PCA
The AMMI statistical model is given as
Where
• is the yield of genotype i in environment j for the kth
replicate,
• is the grand mean,
• is the grand mean, is the genotype i mean deviation
(genotype mean minus grand mean),
17. Statistical analyses ---
Where
• is the environment j mean deviation,
• is the number of singular value decomposition (SVD) axes
retained in the model,
• is the singular value for SVD axis n,
• is the genotype i eigenvector value for IPCA axis n,
• is the environment j eigenvector value for IPCA axis n,
• is GEI residual
• is the error term,
18. 3. Materials ---
Statistical analyses:
The AMMI Stability Value (ASV) was done as
described by Purchase (1997)
Such a measure is essential in order to
quantify and rank genotypes according to their
yield stability,
AMMI Stability Value (ASV) =
19. 4. Results and Discussion
Relative performance of genotypes based
on mean grain yield
1. Mean yield in the tests ranged from 700 -
4278 kg ha-1
• indicating rather divergent conditions for lines,
• expected, in view of geographical differences b/n
the sites of evaluation
2. In terms of mean yield of lines,
• Lines 13 and 7 were the most productive, followed
by lines 12, 8, 4,5 and 11
• The standard check Awash-1 was the least
performing
21. 4. Results and ---
The combined ANOVA indicated
1. Highly significant differences (P<0.01) for
environments, lines and GEI
2. The IPCA axes were also highly significant (P<0.01)
3. Variance components (%) of the SS, ranged from 2%
for lines, 76% for environments and 10% for GEI
This indicated the overwhelming influence that
environments have on the yield performance of navy
bean lines
G x E variation is five times the variation of lines as
main effect
22. Table 3. AMMI ANOVA of grain yield for 16 navy bean lines at
fourteen environments during 2010 – 2011 main crop season
Contribution of
Source DF Sum of Square Mean Square each component
to the total SS (%)
Treatment
223 522867290 2344696** 88
Environments
13 460020679 35386206** 76
Reps within Environment
28 12250577 437521** 2
Line/Genotye
15 9379070 625271** 2
Variety x Environment
195 53467541 274193** 10
Interaction PCA 1
50 27507284 1060538** 51
Interaction PCA 2
46 18340060 773422** 35
Residuals
99 7620196 76972ns 14
Pooled error
420 58068413 138258 10
** and * - stands for 1 and 5% probability levels; ns – non significant
23. 4. Results & Discussion- Stability
To identify the most stable genotypes by
AMMI, the mean of the absolute scores was
obtained for the first two components, weighted
by the percentage of explanation of each
component (weighted mean of absolute scores –
WMAS) for each genotype
Thus, the lower the WMAS value, the lower the
contribution of a genotype to the interaction
and, consequently, the more stable is the
genotype.
24. Table AMMI stability value (ASV) and ranking with the IPCA 1 & 2 scores for the 16 lines evaluated at 14
environments over two years
Line Line name Mean IPCA Score 1 IPCA Score 1 ASV Rank
code
13 ICA Bunsi x S x B 405/1C-C1-1C-87 2435 1.9774 0.4549 2.58 1
7 ICA Bunsi x S x B 405/1C-C1-1C-37 2318 6.0611 18.7532 25.06 13
12 ICA Bunsi x S x B 405/1C-C1-1C-80 2187 -18.0517 -2.8073 23.23 10
8 ICA Bunsi x S x B 405/1C-C1-1C-51 2161 21.7655 3.3169 28.00 15
4 ICA Bunsi x S x B 405/1C-C1-1C-14 2153 19.5516 -1.8071 24.97 12
5 ICA Bunsi x S x B 405/1C-C1-1C-23 2149 -12.9145 0.0620 16.42 4
11 ICA Bunsi x S x B 405/1C-C1-1C-70 2130 8.9726 -6.8554 14.36 3
16 Awash melka 2106 19.4211 8.5139 26.97 14
9 ICA Bunsi x S x B 405/1C-C1-1C-58 2096 -1.8671 -17.5661 22.46 9
6 ICA Bunsi x S x B 405/1C-C1-1C-30 2071 10.2217 -10.1807 18.35 7
2 ICA Bunsi x S x B 405/1C-C1-1C-3 2067 -7.6255 -3.3137 10.57 2
14 ICA Bunsi x S x B 405/1C-C1-1C-88 2061 -9.1268 -9.8879 17.11 6
10 ICA Bunsi x S x B 405/1C-C1-1C-69 2058 -5.7894 14.0161 19.28 8
1 ICA Bunsi x S x B 405/1C-C1-1C-1 2047 -0.6274 -13.4028 17.06 5
3 ICA Bunsi x S x B 405/1C-C1-1C-13 2017 -17.7259 -5.2340 23.50 11
15 Awash - 1 1902 -14.2426 25.9381 37.63 16
Line 13 is High yielding Stable
25. 4. Results & Discussion- Stability
Lin and Binns’s cultivar performance measure (Pi):
As a stability statistic the cultivar performance
measure (Pi) is estimated by the square of
differences between a genotype’s and the
maximum genotype mean at a location, summed and
divided by twice the number of locations
The genotypes with the lowest (Pi) values are
considered the most stable.
From this analysis, the most stable cultivar ranked
first for Pi and for mean yield was Line 13 followed
by line 7 ranked second for Pi and for mean yield.
26. Table Lin & Binns’s (1988a) cultivar performance measure (Pi) for 16 navy bean lines tested at
14 environments, for the years 2010-2011
No Lines Mean Yield Pi(x103) Rank
13 ICA Bunsi x S x B 405/1C-C1-1C-87 2435 28 1
7 ICA Bunsi x S x B 405/1C-C1-1C-37 2318 82 2
12 ICA Bunsi x S x B 405/1C-C1-1C-80 2187 140 3
8 ICA Bunsi x S x B 405/1C-C1-1C-51 2161 170 4
4 ICA Bunsi x S x B 405/1C-C1-1C-14 2153 198 12
5 ICA Bunsi x S x B 405/1C-C1-1C-23 2149 173 5
11 ICA Bunsi x S x B 405/1C-C1-1C-70 2130 217 14
16 Awash melka 2106 197 11
9 ICA Bunsi x S x B 405/1C-C1-1C-58 2096 191 8
6 ICA Bunsi x S x B 405/1C-C1-1C-30 2071 206 13
2 ICA Bunsi x S x B 405/1C-C1-1C-3 2067 179 6
14 ICA Bunsi x S x B 405/1C-C1-1C-88 2061 194 9
10 ICA Bunsi x S x B 405/1C-C1-1C-69 2059 180 7
1 ICA Bunsi x S x B 405/1C-C1-1C-1 2047 196 10
3 ICA Bunsi x S x B 405/1C-C1-1C-13 2017 248 15
15 Awash - 1 1902 367 16
27. 4. Results & Discussion- Stability
Nassar and Hühn, 1987 non-parametric stability analysis
This test is based on the ranks of the genotypes across environments
and gives equal weight to each location or environment.
Genotypes with less change in rank are expected to be more stable.
The mean absolute rank difference (S1) estimates are all possible pair
wise rank differences across locations for each genotype.
The S2 estimates are simply the variances of ranks for each genotype
over environments
For S1, genotypes may be tested for significantly less or more stable
than the average stability/instability.
For the variance of ranks (S2), smaller estimates may indicate relative
stability. Often, S2 has less power for detecting stability than S1
28. Table. Mean absolute rank differences (S1) and variance of ranks (S2) for mean yield of 16
navy bean lines across environments
No Lines Mean Yld S1 S2 Rank
13 ICA Bunsi x S x B 405/1C-C1-1C-87 2435 3.14 3.36 1
7 ICA Bunsi x S x B 405/1C-C1-1C-37 2318 5.43 19.49 2
12 ICA Bunsi x S x B 405/1C-C1-1C-80 2187 8.00 25.38 7
8 ICA Bunsi x S x B 405/1C-C1-1C-51 2161 7.86 17.52 6
4 ICA Bunsi x S x B 405/1C-C1-1C-14 2153 7.36 20.71 3
5 ICA Bunsi x S x B 405/1C-C1-1C-23 2149 7.71 22.68 5
11 ICA Bunsi x S x B 405/1C-C1-1C-70 2130 7.64 26.71 4
16 Awash melka 2106 8.57 22.73 8
9 ICA Bunsi x S x B 405/1C-C1-1C-58 2096 9.43 24.88 10
6 ICA Bunsi x S x B 405/1C-C1-1C-30 2071 9.07 13.15 9
2 ICA Bunsi x S x B 405/1C-C1-1C-3 2067 10.21 13.26 13
14 ICA Bunsi x S x B 405/1C-C1-1C-88 2061 9.71 13.76 11
10 ICA Bunsi x S x B 405/1C-C1-1C-69 2058 10.07 14.07 12
1 ICA Bunsi x S x B 405/1C-C1-1C-1 2047 10.43 11.03 14
3 ICA Bunsi x S x B 405/1C-C1-1C-13 2017 10.50 29.50 15
15 Awash - 1 1902 10.86 20.29 16
29. 4. Results & Discussion- AMMI biplot
The IPCA 1 and IPCA 2 axes explained 51% and 35% of the
total GEI & both are significant at (P<0.01)
By plotting both the lines and the environments on the same
graph, associations b/n lines and the environments can be
seen clearly
The IPCA scores of a genotype is an indication of the
stability or adaptation over environments
The greater the IPCA scores, either negative or positive, (as
it is a relative value), the more specific adapted is a genotype
to certain environments
The more the IPCA scores approximate to zero, the more
stable or adapted the genotype is over all the environments
30. G8
Many lines performed
20 G4
E4G 16 around the mean yld
G13 and G7 are high
E3 E14 yielding lines
10 G6
G11 E1
IPCA 1
G7 E2 G13, G1, and G9 are
E10
E9
G13
stable lines
E7
0 E11 E8 E12 G1
G9
E13
G10
G2
G14
-10
G5
G15
G3 G12
-20
1000 1500 2000 2500 3000 3500 4000
Genotype & Environment means
Figure . IPCA 1 scores for both genotypes and environments plotted
against the mean yield for genotypes and environments
31. E1, E2, E3, E6 and E14 high
20
E2 yielding/favourable envts
G7
G10
E4, E5 and E13 observed
E14 average performance
10 E4
E10
G16
E7, E8, E9, E10, E11 and E12
IPCA 2
E7 E11 E6
E9 G8 are low yielding envts
E5
0 G5 G13
G4
G2 G12
E8 G3
E12 G11
-10 G14
G6
G1 E1
G9
-20
E13
E3
1000 1500 2000 2500 3000 3500 4000
Genotype & Environment means
Figure 2. IPCA 2 scores for both genotypes and environments plotted
against the mean yield for genotypes and environments
32. 4. Results & Discussion- AMMI biplot
Environments spread from the lower yielding
environments in quadrants I and IV to the high yielding
environments in quadrants II and III
High yielding locations are
Melkassa, Haramaya, Alemtena and Jimma
The unfavourable locations for navy bean production are
areas represented by Pawe, Bako, Areka and Sirinka due
to the different biotic and abiotic stresses
The line best adapted to most environments was Line 13
but was also better adapted to the higher
yielding, favourable environments
There also lines with specific adaptation pattern
34. Conclusion
The two high yielding (averaged over environments)
genotypes 13 and 7 could be regarded as a widely
adapted/ stable genotype and having low contributions
to G×E interaction
Genotype 13 combined low absolute IPC1, IPCA2
scores and high yield would be best overall winner
with relatively less variable yield across environments
Favorable test environments should have larger IPCA1
scores (more discriminative) and near zero IPCA2
scores (more representative)