Stability Analysis of Seed Yield of Ethiopian Caraway (Trachyspermum ammi L. Sprague ex Turrill) Genotypes in Multi-Environment Trials

Twelve Ethiopian caraway genotypes were evaluated in eight environments in Ethiopia during 2014 and 2015 under rain fed condition. The field experiment was laid out using randomized complete block design with three replications. The objective of this experiment was to select high yielding and stable Ethiopian caraway genotypes with nonparametric methods. Combined analysis of variance was performed and ten rank-based nonparametric stability parameters were measured. A pooled analysis variance for seed yield indicated that there were highly significant (P<0.001) differences for main effects of genotype, environment and their interaction. The results of principal component analysis revealed that the stability statistics and yield were classified into three groups and related to two contrasting concept of stability. In this study, TOP and rank-sum were found to be useful measures for simultaneously selecting high yield and stable cultivars. These measures selected Gondar 027-2001 and Gondar 023-2000 that gave 2254.7 Kg ha-1 and 2186.1 Kg ha-1 seed yield respectively, as stable and the National Variety Release Committee has released these genotypes as the first Ethiopian caraway varieties for production in 2017.

mean yield. Huehn (1979) and Nassar and Huehn (1987) proposed four nonparametric stability statistics; (1) Si(1)is the mean of the absolute rank differences of a genotype over the n environments; (2) Si(2) is the variance among the ranks over the n environments; (3) Si(3) and Si(6)are the sum of the absolute deviations and sum of squares of rank for each genotype relative to the mean of ranks, respectively. A genotype with lowest value of these statistics is considered to be the most stable. Using ranks of adjusted yield means of genotypes in each environment, Thennarasu (1995) proposed NPi(1), NPi(2), NPi(3) andNPi(4) nonparametric statistics and defined stable genotypes as those whose position in relation to the others remained unaltered in the set of environments assessed. A genotype with lowest value of these statistics is considered most stable. Fox et al. (1990) measure the frequency of each that ranked in the top, middle, and bottom third (TOP) of all tested genotypes across environments. A genotype frequently appeared in the top third for yield rank is considered as the most stable. Kang (1988) nonparametric stability parameter applies both yield rank and Shukla's Stability variance (Shukla, 1972). The genotypes that score lowest rank-sum are the most preferred ones. Studying the relationship among nonparametric stability parameters using rank correlations is pertinent to identify the appropriate stability parameters (Mohammadi et al., 2009) and to relate stability parameters with statistic and dynamic concepts of stability (Becker and Leon, 1988). The objectives of the present study were to perform yield stability analysis using non parametric parameters, to study the relationships among stability parameters, and to select high yielding and stable Ethiopian caraway varieties.

Materials and methods
Planting materials and test environments The Ethiopian caraway landraces were originally collected from different Ethiopian caraway growing areas of Amhara region and then morphologically characterized. Then twelve accessions were selected from previous preliminary yield trials that had been conducted in consecutive years. Since there was no registered variety, standard check was not used. Twelve Ethiopian caraway accessions were tested in 8 environments (year and location combinations) during 2014 and 2015 in Ethiopia under rain fed condition. The testing environments and genotypes are described in Table 1 and 3, respectively.
Experimental layout and field management In all environments, the field experiments were laid out using randomized complete block design with three replications. Each genotype was planted on a plot size of 1.8 m long with five rows of 30 cm apart. Thinning was done to have 15 cm space between plants. Nitrogen was applied at rate of 45 kg ha-1at planting and P2O5 was applied at rate of 30 kg ha-1, half at planting and half before flowering. At Kulumsa, supplementary furrow irrigation was provided at flowering stage for two times on ten days interval. Seed yield data collected from central three rows were converted into kg ha -1 .

Statistical analysis
Seed yield data were subjected to combined analysis of variance (ANOVA) using GLM procedure of SAS statistical package (SAS, 2002). Variance homogeneity was tested using Barttlet test and the variance across eight environments varied significantly. Hence, ten nonparametric stability statistics, Si(1), Si(2), Si(3) and Si(6) (Huehn, 1979;Huehn and Nassar, 1987), NPi(1), NPi(2),NPi(3) and NPi(4) (Thennarasu, 1995), rank-sum (RS) (Kang, 1988), TOP (Fox et al, 1990) were computed using R statistical package called "phenability" (Branco, 2015). Spearman rank correlation coefficients among nonparametric stability parameters and with mean yield were produced using SAS statistical package (Steel and Torrie, 1980;SAS, 2002). To further study the association among stability parameters and with yield, principal component analysis (PCA) was performed using rank correlation matrix in Genstat (Genstat, 2015).  (Table 2) for seed yield of 12 Ethiopian caraway genotypes studied across eight environments indicated that there were significant (P<0.001) differences for the effect of genotype (G), environment (E) and G x E interaction. Out of total variation, environment contributed maximum variance (79.3%), indicating its largest effect on seed yield than the effect of genotype (1.11%) and G x E interaction (9.24%). In multi environment trials, variance of environment is known to be largest (80%) while G x E interaction and genotype are usually small (Yan and Kang, 2003). The largest environmental variance might be resulted from the agro-ecological variation among test locations (Table 1). However, the most relevant for genotype evaluation are the genotypic and G x E interaction effects and environment effect is usually ignored (Yan and Kang, 2003). In this study, the G x E effect exceeded the genotype effect 8 times, showing significant G x E interaction effect suggesting the possible presence of different mega environments with different top-yielding genotypes and the genotypes performed variably across environments (Yan and Kang, 2003). Magari and Kang (1993) and Kang (1990) demonstrated that significant interaction of genotype by environment creates trouble in selecting stable cultivars. Hence, selecting superior genotypes depending on stability and yield performance would be appropriate. Table 2: Combined analysis of variance (ANOVA) table CV=12.9, Mean= 1995.4, *** Significant at the 0.001 probability level. ns Non-Significant at the 0.05 probability Seed yield performance The average seed yield of Ethiopian caraway genotypes across environments ranged from 1762.9 kg ha-1for genotype Adet 12-2000 (G2) to 2254.7 kg ha-1for genotype Gondar 027-2001 (G4) ( Table 3). The maximum environmental mean seed yield was obtained from Takusa-2014 (3282.8 kg ha-1). The minimum environmental mean seed yield was obtained from Assosa-2015 (841.6 kg ha-1) and Chefa-2015 (857.1 kg ha-1) ( Table 3). Heidaria et al. (2016) found that seed yield of Ajowan (Trachyspermum ammi L.) accessions collected in Iran ranged from 1432.1 kg ha-1to 4539.2 kg ha-1.  Table 3:Combined mean seed yield of 12 Ethiopian caraway genotypes studied in 8 environments Nonparametric stability analysis The value of ten nonparametric stability statistics using seed yield and their rank based on the value were presented in Table 4 and 5, respectively. The significant tests, Z1 and Z2, values for Si(1) and Si (2) were calculated using ranks of adjusted data for each genotype and added over to get Z1 and Z2 sum, respectively (Table 4) (Nassar and Huehn, 1987). The critical value of χ 2(21.03) (P<0.05, df=11) exceeded both Z sum values, showing that there were no significant differences in rank stability among genotypes. Besides, none of any genotype had higher Z value than the critical value χ 2 (8.2)(P<0.05, df=1), indicating all genotypes performed significantly stable, relative to others. G1 and G12were considered as the most stable genotype since they had the minimum value of both Nassar and Huehn's (1987) statistics, Si (1) and Si(2). The highest yielding genotype, G4, ranked the third most stable according to both stability statistics (Table 5). G5 showed maximum value of Si(1) and Si(2), indicating to be the most unstable genotype. Si(3) and Si(6) stability statistics combine yield and stability performance and the lowest values of these statistics indicate the most stable genotype (Huehn, 1979). Both Si (3) and Si (6) statistics identified G2 and G6 as the most stable and G9 as the most unstable genotypes. Thennarasu's (1995) nonparametric statistics, NPi(1), NPi(2), NPi (3) andNPi (4), consider a genotype with lowest value of these statistics as the most stable. According to NPi(1), G1 and G12 scored the lowest value, hence they were the most stable genotypes and G5 and G11 were the most unstable. G1showed small value of NPi(2), indicating to be the most stable, followed by G8 and G12. Like NPi(1), NPi(3) selected G1and G12 as the most stable. Statistics NPi(4) identified G6, G3 and G2 as the most stable genotypes. Three of Thennarasu's (1995) statistics, NPi(2), NPi(3) and NPi(4) indicated that the highest yielding genotypes, G4 and G11 were the most unstable (Table 5). Similarly, Yong-Jian et al. (2010) found that NPi(2), NPi(3) and NPi(4) identified higher yielding genotypes as unstable in maize multi-environment trials. TOP stability measure chose the best yielding genotypes G4 and G11 as the most stable followed by G9 and G10, since they ranked in the top third in the majority of environments (Fox et al.,1990) (Table 5 and 6). According to Temesgen et al. (2015), TOP measure identified high yielding genotypes to be the most stable in faba bean. The undesirable genotypes identified by TOP measure were G6. According to Kang (1988), nonparametric stability statistics, genotypes that score lowest rank-sum (RS) value are the most preferred ones. Two best yielding (G4 and G11) and three relatively lower yielding (G10, G1, and G5) genotypes had lower rank-sum (RS) and were considered as the most stable (Table 5 and 6). G3, G6, and G8 were identified to be undesirable genotypes by the rank-sum statistic. Table 5: Ranks of 12 Ethiopian caraway genotypes using 10 different nonparametric stability methods

Relationships among nonparametric stability statistics
The result of spearman's rank correlation shown on Table 6 indicates that only TOP parameter was significantly (p<0.01) and positively correlated with mean yield, indicating TOP could be importantly used for selecting genotypes with high wider adaptability in Ethiopian caraway. Strong positive correlation of mean yield with TOP has been also reported in lentil (Sabaghnia et al., 2006), bread wheat (Gebru and Abay, 2013), faba bean (Temesgen et al., 2015) and durum wheat Amri, 2008, Kaya andTurkoz, 2016) genotypes. Mean yield and TOP measure were significantly and negatively correlated with NPi(2), NPi(3), NPi(4) and Si(6) and these four statistics were correlated significantly and positively to each other (Table 6). Similarly Sabaghnia et al. (2006) and Mohammadi et al. (2009) have reported that yield and TOP measure were negatively correlated with NPi(2), NPi(3), NPi(4) and Si(6) in lentil and barley genotypes, respectively. These four statistics were correlated positively to each other in lentil and barley genotypes (Sabaghnia et al., 2006;Mohammadi et al., 2009). In this study, NPi(2), NPi(3), and NPi(4) statistics were not significantly correlated with NPi(1). Similarly, absence of significant association of NPi(1) with the remaining NPis' was also reported in lentil and maize genotypes (Sabaghnia et al,2006;Yong-Jian et al., 2010). There were significantly (p<0.01) positive rank correlation among Si(1), Si(2) and NPi (1) but these statistics were not correlated with mean yield. Sabaghnia et al. (2006), Mohammadi and Amri (2008), Mohammadi et al. (2009), andTurkoz (2016) have reported similar result in multi-environment trial of genotypes of different crops. In agreement with the current result, significantly positive correlation between Si(3) and Si(6) and significantly negative correlation with mean yield were indicated by Mohammadi et al. (2007a), Segherloo et al. (2008), Shah et al. (2009), andTemesgen et al. (2015). RS was not significantly correlated with mean yield and TOP. Mohammadi and Amri, (2008) reported similar result in durum wheat. According to Mohammadi et al. (2009), RS was not significantly correlated with TOP in genotypes of four studied crops. Principal component analysis Principal component analysis (PCA) was computed to further investigate the relationship among nonparametric stability parameters and with mean yield. Figure 1 shows a bi plot graph of PCA1 against PCA2. The first two principal components described 80.76% (54.55% and 26.21% by PCA1 and PCA2, respectively) of the original variance. The bi plot of the PCA1 and PCA2 divided stability parameters and mean yield into three groups ( Figure  1).
In this study, PCA1 distinguished stability parameters and mean yield based on two contrasting concepts of stability: the static (biological) and dynamic (agronomic) concepts. The concept of both stabilities is explained in detail by Becker (1981) and Becker and Leon (1988). Regarding to static stability/biological concept, a stable genotype performs constant yield regardless of environmental variations. On the contrary, according to dynamic/agronomic stability, the yield performance of a stable genotype responds to environmental variation (Becker, 1981;Becker and Leon, 1988).
Parameters clustered in Group 1 (TOP and RS) had positive correlation among themselves and with mean yield in Ethiopian caraway genotypes (Table 6). These stability parameters, TOP and RS, are related to dynamic/agronomic stability concept. Several previous studies indicated that TOP and RS are positively associated with mean yield and are related to the dynamic concept of stability (Sabaghnia et al., 2006;Mohammadi and Amri, 2008;Flores et al., 1998;Mohammadi and Ahmed, 2013;Akcura and Kaya, 2008). Therefore, these parameters could be recommended as useful measures for cultivar selection; according to Becker (1981) and Mekbib (2002), most plant breeders are interested in selecting for high yield and stability, simultaneously.
Hence, Top, and RS could be used to select high yielding Ethiopian caraway genotypes stable to wide range of environments. In this study, Gondar 027-2001 (G4) and Gondar 023-2000 (G11) were found to be high yielding and stable based on the TOP and RM parameters that are related to dynamic concept of stability. Besides, genotype Gondar 027-2001 (G4) also ranked as third most stable based on two static concept statistics, Si(1) and NPi(1) that did not correlate with yield. According to Becker and Leon (1988), in dynamic concept of stability, stable genotype should not necessarily have a constant performance across environments. Hence, parameters associated with dynamic concept of stability could be used to identify cultivars for high potential environments. However, as pointed out by Roostaei et al. (2014), these stability parameters might drop low general adaptable but high specific adaptable genotypes.
Stability parameters clustered in group 2 (Si(1), Si(2) and NPi(1))were correlated significantly (p<0.01) and positively to each other. Si(1), Si(2) and NPi(1) statistics ranked genotypes for stability similarly suggesting one of these parameters can be used as an alternative to the other parameters. Statistics Si(1), Si(2) and NPi(1) (group 2) matched with static/biological concept of stability and were not significantly correlated with mean yield, the group 1 (dynamic stability) and group 3 (static stability) statistics. In agreement with the current result, Sabaghnia et al. (2006), Kaya and Turkoz (2016) and Mohammadi and Amri (2008) indicated that Si(1), Si(2) and NPi (1) were not correlated with yield and related to static/biological concept of stability. Previous research (Nassar and Huehn, 1987;Mohammadi et al., 2007b) have also shown that Si(1)and Si(2)defined stability in the sense of homeostasis and are related to the static (biological) concept of stability. Group 2 statistics (Si(1), Si(2) and NPi(1)) were influenced simultaneously by both mean yield and stability. Therefore, as also reported by Sabaghnia et al. (2006), Si(1), Si(2) and NPi(1) parameters could be used as compromise methods to select genotypes with moderate yield and high stability (Sabaghnia et al., 2006).

Conclusion
Plant breeders and farmers prefer to select high yielding genotypes with good stability. So, both yield and stability should be simultaneously considered in genotype selection. Therefore, the use of agronomic or dynamic concept of stability is better. The present study indicated that group 1 nonparametric statistics, TOP (proportion of environments at which the genotype occurred in the top third) and rank-sum (sum of ranks of mean yield and Shukla's stability variance), positively correlated with mean yield and associated with the dynamic concept of stability. This study indicated that TOP and RS nonparametric statistics could be used to select genotypes for both high yielding and better stability simultaneously, in future Ethiopian caraway breeding program. Based on TOP and RS stability measures, Gondar 027-2001 (G4) and Gondar 023-2000 (G11) were high yielding and stable genotypes. Hence, National Variety Releasing Committee under the Ministry of Agriculture and Natural Resources of Ethiopia approved the release of 027-2001 (G4) and Gondar 023-2000 (G11) for production and named as "Dembia-01" and "Takusa-01" respectively (MoANR, 2017).