INTRODUCTION
Safflower (Carthamus tinctorius L.) is a multipurpose
plant cultivated since ancient times not only for the dye contained in
its flowers, the oil in its achenes and its medicinal properties, but
also for the ornamental value of its colorful inflorescence. In 2005,
there area of safflower production in the world was estimated to be about
814,000 ha (FAO, 2006). More than 20 countries grow safflower and Mexico
and India produce over half, with 212,765 and 210,000 Million tonnes,
respectively. There are now proactive efforts to create and develop area
for important oilseed crops; in line with this government is now encouraging
safflower cultivation for edible oil purpose. In the last few years safflower
area has increased and was 15,000 ha in 20052006. However, genotype x
environment (GE) interaction is a major importance to the plant breeders
for developing safflower cultivars in rainfed conditions of Iran. GE
interaction can be an outcome of genotype rank changes from one environment
to another, a difference in scale among environments, or a combination
of these phenomena (Mohammadi et al., 2007). If relative performances
of the entries grown in different environments are highly different, then
GE interaction becomes a major challenging factor to crop breeding programs
(Zobel and Talbert, 1984). In such cases, the breeder is faced either
with developing specific breeding populations for each environment and/or
with selecting genotypes that generally perform well across many environments
(Isik and Kleinschmit, 2005). Some genotypes are adapted to a broad range
of environmental conditions, while others are more limited in their potential
distribution and have specific adaptation.
The patterns of genotypes response to environmental change
can be summarized in the form of the regression of genotype means on environmental
index. The joint regression stability analyses of Finlay and Wilkinson
(1963) and Eberhart and Russell (1966) have been widely used. Genotypes
with low coefficients (b_{i}) either show specific adaptation
or little variation with environment; those genotypes with high slopes
(b_{i}) are more responsive to improvements in the environment
and generally more adapted to favorable environments (Lin et al.,
1986; Simmonds, 1991). When different genotypes are tested in a range
of specific environments, generally the contribution of each genotype
(ecovalence) to the total interaction sum of squares is estimated (Wricke,
1962; Becker and Leon, 1988; Karlsson et al., 2001; Isik and Kleinschmit,
2003).
The proportion of sites at which the genotype occurred
in the top, middle and bottom third of the ranks was computed to form
the parameters TOP_{i}, MID_{i} and LOW_{i}, respectively
(Fox et al., 1990). A genotype that occurred mostly in the top
third (high value of TOP_{i}) was considered to be a widely adapted
genotype. Some other univariate parameters are: environmental variance
(S^{2}_{i}) (Roemer, 1917), coefficient of determination
(R^{2}_{i}) proposed by Pinthus (1973) and Francis and
Kannenberg`s (1978) coefficient of variation (CV_{i}) which suggested
for each genotype. More recently, Purchase (1997) developed the AMMI Stability
Value (ASV) based on the AMMI model`s IPCA1 and IPCA2 (Interaction Principle
Components axes 1 and 2, respectively) scores for each genotype (ASV_{i}).
The goals of this study were to identify safflower genotypes
by simultaneously selecting for yield performance and stability, to estimate
the contribution of each test environment to total GE interaction and
to study interrelationships among studied stability parameters.
MATERIALS AND METHODS
Data collection: This study was carried out with
16 advanced spring safflower genotypes, PI537589 (G1), Syrian (G2), PI537636
(G3), CW4440 (G4), Lesaf (G5), Cyprus bregon (G6), CW74 (G7), Kino76
(G8), S541 (G9), PI250536 (G10), PI250537 (G11), Hartman (G12), Gila
(G13), PI537636s (G15), PI198290 (G16) and Dinçer (G17) and
Mahali Isfahan (G14) as standard check at nine locations i.e., Sararood,
Maragheh, Gachsaran, ShirvaneKhorasan, Ghamlo, Khodabandeh, Ardabil,
Shirvan Chardavel, Khoram Abad, which are representative of different
safflower growing areas under rainfed conditions of Iran in the three
growing seasons of 200304 to 200506 (data of location of Khoram Abad
in 20052006 was not available). The descriptive of the trial sites is
shown in Table 1. At each environment the genotypes
were planted in a randomized complete block design with three replications.
Sowing was done by hand. Plot size was 6 m^{2} (4 m length, 5
rows and 30 cm between adjacent rows). Plants were spaced 10 cm apart
within rows. The area harvested was 3.6 m^{2}, however, only the
middle three rows were harvested. Fertilizer application was 50 kg N ha^{1}
and 50 kg P_{2}O_{5} ha^{1} at planting. Seed
yield (kg ha^{1}) was obtained by converting the grain yields
obtained from plots to hectare. In this study some stability parameters
were applied to the data chosen so that they cover a wide range of philosophies
in stability analysis.
The method of Finlay and Wilkinson (1963) was used to
estimate regression coefficient (b_{i}). The stability of grain
yield was calculated by regression the mean yields of individual genotypes
on environmental index. Coefficient of determination (R^{2}_{i})
(Pinthus, 1973) and environmental variance (S^{2}_{i})
(Roemer, 1917) was also
Table 1: 
Description of the experimental sites and their overall
agroclimatic conditions, like total annual rain fall and average
minimum and maximum temperature 

computed, where a genotype with the maximum R^{2}_{i}
value and minimum variance is considered to be stable. The stratified
ranking technique of Fox et al. (1990) was considered to form the
measures TOP_{i}, MID_{i} and LOW_{i}. A genotype
that occurred mostly in the top third (high value of TOP) was considered
to be a widely adapted genotype. The stability was measured by combining
use of coefficient of variation (CV_{i}) and mean yield (Francis
and Kannenberg, 1978). Ecovalence (W^{2}_{i}) as suggested
by Wricke (1962) was computed to further describe stability. A low W^{2}_{i}
value indicates high relative stability. The AMMI Stability Value (ASV)
(Purchase, 1997) based on the AMMI model`s IPCA1 and IPCA2 scores for
each genotype. ASV is in effect the distance from the coordinate point
to the origin in a two dimensional scattergram of IPCA1 scores against
IPCA2 scores. The largest the IPCA scores, either negative or positive,
the more specific adapted a genotype is to certain environments, the smallest
IPCA scores, there more stable the genotype is over all environments sampled.
RESULTS
AMMI analysis: The analysis of variance (additive
main effects) showed significant effects for genotype (G), environment
(E) and GE interaction (Table 2). These results showed
that 83.78% of the treatment Sum of Squares (SS) (G+E+GE) was attributable
to environment effects, only 1.37 and 14.85% to genotype and GE interaction
effects, respectively. Results from analysis of multiplicative effects
also showed that the first Interaction Principle Component Axis (IPCA1)
captured 27.34% of the interaction SS in 10.0% of the interaction degrees
of freedom (df). Similarly, the IPCA2, IPCA3 and IPCA4 explained a further
15.42, 13.61 and 9.9% of the GE interaction SS, respectively. In total,
AMMI2 model (G+E+IPCA1 and IPCA2) contained 91.51% of the treatment SS,
while the residual contained only 8.49%. These results indicate that the
AMMI model fits the data well and justifies the use of AMMI2.
Analysis of stability performance: The genotypes
showed significant differences in grain yield. Taking mean yield as a
first parameter for evaluating the genotypes, G1, G15, G2, G5 and G4 gave
the best mean yields while G7, G12, G14, G8 and G16 had the lowest mean
yields across environments (Table 3, 4).
The IPCA scores of a genotype in the AMMI analysis are an indicator of
the stability of a genotype over environments (Purchase, 1997). The lowest
IPCA1 was observed for the genotypes G7 followed by G6 and G3 and IPCA2
was in the lowest for the genotypes G4, G13 and G7 (Table
3) and the ranks of genotypes according to this parameter are given
in
Table 2: 
Additive main effects and multiplicative interactions
analysis of variance for grain yield of the 17 genotypes in 26 environments 

Note: The block source of variation refers to blocks
within environments, **: Significant at 1% level of probability, ^{@}:
The data of one the environments (Location of Khoram Abad in 200506)
was not available 
Table 4. According to IPCA1 and 2,
G7 was the highest stable genotype with the mean yield (647 kg ha^{1})
lower than grand mean (700 kg ha^{1}). The highest IPCA1 was
belonging to G17 followed by G13 and G2 with the higher grain yield than
grand mean and the lowest IPCA2 was belonging to G1 (773 kg ha^{1})
followed by G15 (733 kg ha^{1}) which had the highest mean yield.
The AMMI stability value (ASV_{i}) confirms the results of IPCA
1 and 2 scores. However, ASV_{i} ranked the genotype G7 with the
lowest ASV_{i}, as the most stable genotype, although it had the
lowest yield performance (647 kg ha^{1}). Corresponding to ASV_{i}
the G1 was instable although had the highest yield performance. G17 was
as the most instable genotype but of high adapted to the testing environments.
In keeping Wricke`s (1962) stability parameter, W^{2}i,
the genotypes G13 followed by G4 and G3 with the lowest ecovalence and
were considered to be stable which being responsible for 1.5, 1.7 and
1.8% of the total interaction sum of squares, respectively, whereas the
G17 followed by G1 with the highest W^{2}_{i} were instable
and had the most contribution to GE interaction. The regression coefficients
for the seventeen genotypes examined was ranged from 0.70 to 1.02. Corresponding
to Finlay and Wilkinson`s (1963) method, the genotypes G8, G10 and G16
had coefficient regression (bi) value equal to one and the genotypes G3,
G7, G12, G9 and G13 with values closer to one were more stable. The genotypes
with the lowest bi (especially, G2, G6 and G15) were adapted to marginal
environments.
The coefficient of determination (R^{2}_{i})
represent agronomic stability (Becker, 1981), which is the predictability
of estimated response (R^{2} = 1.0). The predictability of genotypes
for the yield was varied. The values ranged from 0.82 (for G1) to 0.98
(for G4) which indicated that 82.0 to 98.0% of the mean yield variation
Table 3: 
Mean yield and genotypic stability parameters for 17
safflower genotypes across 26 environments 

^{a}: Contribution of each genotype to GE interaction 
Table 4: 
Ranks of 17 genotypes based on mean yield and parametric
measures derived from yield across 26 environments 

Table 5: 
Spearman`s rank correlation between mean yield and
stability parameters 

*, **: Significant at 5 and 1% levels of probability,
respectively 
was explained by genotype response across environments.
Roemer (1917) stability index, S^{2}_{i}, which describes
biological stability (Becker, 1981), quantitatively reflects the yield
of a genotype in all environments. Therefore genotypes such as G9, G10
and G16 have low biological stability unlike the genotypes G17, G1 and
G6 with the highest S^{2}_{i} (Table 3).
Corresponding to parameter of Fox et al. (1990),
G1 was an adapted genotype, because it ranked in the top third of genotypes
in a high percentage of environments (high top value, 65%) and was followed
by G9 (62%) and G11 (58%) (Table 3). The undesirable
genotypes identified by this method were G12 and G7. According to Francis
and Kannenberg`s (1978) stability parameter (CV_{i}), the genotypes
G9, G10 and G13 were considered to be stable genotypes. These genotypes
with the lowest CV_{i} were medium in yield. The genotypes G17,
G2 and G6 with the highest CV_{i} values had high yield performance.
Interrelationship among stability parameters:
The ranks of 17 genotypes and 26 environments after applying the method
stability analysis were used to rank correlation (the ranks of genotypes
and environments are not shown). Spearman`s rank correlations coefficient
among genotypic mean yields with the parameters are shown in Table
5. The means of genotype yield were positive correlated with the genotypic
parameters of TOP_{i} (p<0.01) and bi (p<0.05) but in negative
correlated with S^{2}_{i} and IPCA2i (p<0.05) (Table
5). The bi was strongly negative correlated with S^{2}_{i},
CV_{i} and W^{2}_{i}.
DISCUSSION
Analysis of GE interaction and estimation of phenotypic
yield stability have been widely studied and several methods were proposed
for its estimation (Wricke, 1962; Eberhart and Russell, 1966; Pinthus,
1973; Francis and Kannenberg, 1978; Lin et al., 1986; Becker and
Leon, 1988; Purchase, 1997). One of the reasons for growing genotypes
in a range of environments is to estimate their phenotypic stability because
of the increasing grower demands for stable varieties especially in areas
where climatic conditions are highly unpredictable (Ceccarelli, 1994).
In breeding programs genotypes are tested in numbers
of environments. Environmental variations seemed to be of important to
in determining performance, so, evaluation based on several years and
locations is a good strategy to be pursued in breeding programs (Mohammadi
and Amri, 2007). Farmers in developing countries which use no or limited
inputs or growing safflower under harsh and unpredictable environments,
will need stable varieties. In these cases genotypes with good performance
and stability should be recommended. Stability performance of genotypes
is the most important factor under rainfed conditions in Iran, where
environmental conditions vary considerably (Mohammadi and Amri, 2007).
In major problem of safflower improvement program in Iran has been the
lake of genotypes consistently perform well across different safflower
growing environments. Hence, the development of high yielding genotypes
and information multilocation stable performance are a paramount importance
in Iran where environments vary greatly within short distances.
However, several of stability measures that have been
used in this study quantified stability of genotypes with respected to
mean yield, stability and the best combination of them. Most of were closely
related in sorting out the relative stability of the evaluated safflower
genotypes. Some deviations were, however, observed specially for the genotype
superiority measure. Purchase (1997) and Adugna and Labuschagne (2003)
also reported similar results, indicating that it was more of a performance
measurement than a yardstick for stability of genotypes across environments.
In summary, according to stability parameters the genotypes G9, G10 and
G11 with a good combination of yield and stability can be selected, whereas
the genotypes G1 and G17 as unstable ones with high yield performance.
The remaining genotypes were intermediate between these two groups.
In conclusion, several of stability statistics that have
been used in this study quantified stability of genotypes with respected
to yield, stability and both of them. Therefore, both of yield and stability
should be considered simultaneously to exploit the useful effect of GE
interaction and to make selection of the genotypes more precise and refined.
ACKNOWLEDGMENTS
This research was part of a Regional Safflower Research
Project of Drayland Agricultural Research Institute (DARI) of Iran and
sponsored by the Agricultural Research and Education Organization (AREO).
We thank all members of the project for any contribution they may have
made towards this study.