Univariate stability analysis and relationship among parameters for grain yield of striga resistant sorghum [Sorghum bicolor (L.) Moench] hybrids in Ethiopia

Sorghum (Sorghum bicolor) known as a Camel crop of cereals, is among the dominant staple food grains for the majority of Ethiopians. Forty nine sorghum genotypes (hybrids + open pollinated varieties) were tested at fi ve locations in a simple lattice design with two replications during the 2016 main cropping season. The objectives of this study were to determine yield stability using univariate methods and to assess the association among stability parameters of striga resistant sorghum genotypes in the dry lowland areas of Ethiopia. The result of the combined analysis of variance for grain yield revealed highly signifi cant (P≤0.001) difference among Environment (E), Genotype (G) and Genotype × Environment Interaction (GEI). Based on the combined ANOVA over locations, the mean grain yield of environments ranged from 588 kg ha-1 in Humera to 4508 kg ha-1 in Sheraro. The highest yield was obtained from ESH-1 (3278 kg ha-1), while the lowest was from K5136 (735 kg ha-1) and the average grain yield of genotypes was 2184 kg ha-1. Different stability models were used in measuring of genotype stability such as AMMI Stability Value (ASV), Yield Stability Index (YSI), coeffi cient of regression (bi) and deviation from regression (S2di). Yield was signifi cantly correlated with bi (0.91), r2 (0.55) and ASV (-0.56), while it was not correlated with S2di (-0.26). The non-signifi cant correlation among yield and stability statistics indicated that, stability statistics provide information that can not be collected from average yield. The high positive correlation among mean grain yield and stability parameters is expected as the values of these parameters were higher for high yielding genotypes and the vice versa. Highly correlated stability parameters indicate that they can measure stability similarly. However, there were inconsistencies with the univariate stability parameters used, which created uncertainty to select or recommend the stable genotypes. Therefore, as the data is from one year, it is necessary to repeat the experiment at least for one more year across diverse dry lowland areas of Ethiopia. Received: 24 June, 2021 Accepted: 09 July, 2021 Published: 10 July, 2021 *Corresponding author: Fantaye Belay, Abergelle Agricultural Research Center, P.O. Box 44, Abi-Adi, Ethiopia, Tel: +251912725888; Fax: +251 344 461 035; E-mail:


Introduction
Sorghum [Sorghum bicolor (L.) Moench] is naturally selfpollinated monocotyledon crop plant with the degree of spontaneous cross pollination, in some cases, reaching up to 30%, depending on panicle type [1]. It is a staple crop for more than 500 million people in 30 sub-Saharan Africa and Asian countries [2]. In Ethiopia, sorghum is produced by fi ve million small holder farmers and its production is estimated to be four million metric tons from nearly two million hectares of land, giving the potential average grain yield of around two tons per hectare. It is ranked third in area coverage and fourth in total production [3]. However, low yields of sorghum have been recorded due to a number of biotic and abiotic constraints.
Sorghum production constraints vary from region to region within Ethiopia; but, drought and striga are reported to be important sorghum production constraints in the north and northeastern parts of the country [4].

Research Article
Univariate stability analysis and relationship among parameters for grain yield of striga resistant sorghum

[Sorghum bicolor (L.) Moench] hybrids in Ethiopia
Stiriga hermonthica, the dominant striga species, is the most severe in the highly degraded north, northwestern and eastern parts of the country, viz. Tigray, Wollo, Gonder, Gojam, North Shewa and Hararghe [5]. Where soil fertility (nutrient defi ciency) and moisture stress are limiting factors, i.e. striga is rapidly expanding in areas where the soil has low fertility and drought is frequent. Nationally, striga causes annual yield loss as high as 65-70% and, at times, leaves plot uncultivated [6].
Many researchers [7,8] have reported variability in sorghum responses to striga infestation. The presence of a wide range of variability in striga resistant and/or drought tolerance traits among sorghum genotypes suggests an opportunity to develop high yielding and resistant/tolerant genotypes through hybridization [9]. In order to address the constraints affecting sorghum, and increase its production, the National Agricultural Research Systems (NARS) in collaboration with international research centers like, ICRISAT and Purdue University are developing hybrid sorghums.
The numerous importances attached to sorghum hybrids stems from the fact that there has been a yield advantage of sorghum hybrids whenever they are compared to the improved and landrace cultivars, commonly in order of 20 to 60% [10]. Sorghum hybrids have been shown to yield 15 to 41% higher than open pollinated varieties under small holder conditions in India and West Africa [11,12]. Reports from research has shown that sorghum hybrids holds a lot of importance and appear to be more reliable than inbred varieties in erratic environments, typically of sorghum growing regions in the semi-arid tropics [13].
One of the importance attached to sorghum hybrids whenever they are compared to the open pollinated and landrace cultivars, increase the yield in order of 20 to 60% [10]. Beside yield superiority over open-pollinated varieties, hybrids are more stable across different environments [14] and more tolerant to moisture stress. In Ethiopia, hybrids give 27-30% more grain yield advantage as compared to check varieties and proved to be early maturing than their parental lines [6,15,16].
The yield advantage in sorghum hybrid is due to the complementarity effect of the two inbred lines on the F1 hybrid [17]. It is thus presumed that inbred lines that have striga resistant genes complement each other and the F1 hybrids express superiority in reaction to striga and could give better yield. Abebe, et al. [18] also reported that most resistant sorghum hybrids produced consistently higher grain yields under S.hermonthica infestation, supported fewer emerged parasites, and less sustained minimal parasite damage symptoms across locations. However, there is no information on yield stability of striga resistant sorghum hybrids in Ethiopia. Therefore, the specifi c objectives of the study were to determine yield stability using univariate methods and to assess the association among commonly used stability parameters for striga resistant sorghum hybrids in dry lowland areas of Ethiopia.

Description of the study sites
The fi eld experiment was conducted during the 2016 main cropping season at fi ve locations (Sheraro, Kobo, Mehoni, Fedis and Humera), representing the dry lowland areas of Ethiopia located in the altitude range of 609 -1600 meter above sea level (m.a.s.l), where sorghum is widely grown. The detailed agro-ecological features of the locations are presented in Table   1, Figure 1.

Experimental materials
Breeding materials comprised of 49

Experimental design and crop management
The trial was laid out using a 7x7 lattice design with two replications in each location. Each plot consisted of two rows of 5 m length with 0.75 m and 0.20 m, between rows and plants, respectively. All plots were fertilized uniformly with 100 kg ha -1 Di-ammonium Phosphate (DAP) and 50kg ha -1 Urea. Full dose of DAP and half of urea were applied at the time of planting and the remaining half was side dressed at knee height stage of the crop. All of the other agronomic management practices were applied as required at all locations as per the recommendations for sorghum in dry lowland areas of Ethiopia.

Data collection
Data were collected both on plot and plant basis, based on the descriptors list for sorghum (IBPGR/ICRISAT, 1993).
Phenological data (days to emergence, fl owering, grain fi lling period and maturity date), morphological data (plant height and panicle length), and yield and yield related traits (grain yield and thousand grain weight) were collected.

Data collected on plant basis
From the two rows fi ve plants were selected randomly and tagged to collect the morphological data such as, plant height and panicle length. The detail of the data collection for each trait was carried out as follows: Plant height (PH): was determined from the average height of fi ve plants in cm from ground level to the tip of the panicle (at physiological maturity).
Panicle length: was measured (cm) from the base of the panicle to the tip from fi ve randomly selected plants per plot at maturity.

Data collected on plot basis
Days to 50% seedling emergence: The number of days from the date of sowing to the date at which 50% of the seedlings in a plot were emerged. Grain yield (kg ha -1 ): The panicles from the two rows of each plot were threshed, cleaned and adjusted to standard moisture level at 12.5% and weighted to get the grain yield per plot in grams and converted to kg ha -1 for analysis.
Thousand grain weight: The weight of 1000 randomly sampled grains from each plot was measured in grams using sensitive balance and adjusted at 12.5% moisture content.

Data analyses
Homogeneity of residual variances was tested prior to analysis over locations using Bartlett's tests [19]. Analysis of variance for each environment, combined analysis of variance over environments, correlation coeffi cient among stability parameters and agronomic traits were computed using GenStat 18 th edition (2016. Coeffi cient of regression (bi) and deviation from regression (S 2 di) stability parameters were also analyzed using SPAR 2.0 software.

Individaul and combined ANOVA
As the error variance was homogenous for all traits continued to combined analysis of variance from the mean data of all environments to detect the presence of GEI. Genotypes were assumed to be fi xed and environment effects were treated as random. Genotype by environment interaction was quantifi ed using pooled analysis of variance, which partitions the total variance into its component parts (genotype, environment, genotype x environment interaction and pooled error). Mean separations for the treatment means having signifi cant differences at 5% probability levels was done using Duncan's Multiple Range Test (DMRT) comparison procedure. GenStat 16 th edition (2016) statistical software was used for statistical analyses. The relative effi ciency of the simple lattice design over Randomized Complete Block Design (RCBD) was checked. For most of the yield and yield related traits RCBD was found to be more effi cient than that of the lattice design. The analysis of variance for each location and combined analysis of variance over locations was used as suggested by Gomez and Gomez (1984). The model employed in the analysis was; Yijk = μ + Gi + Ej + Bk + GEij + εijk where: Yijk is the observed mean of the ith genotype (Gi) in the jth environment (Ej), in the kth block (Bk); μ is the overall mean; Gi is effect of the ith genotype; Ej is effect of the jth

Eberhart and Russell's stability analysis
Eberhart and Russell [20] procedure involves the use of joint linear regression where the yield of each genotype is regressed on the environmental mean yield. Then, the behavior of the genotype was assessed by the model: Y ij = μ i + i I j + ij using Spar 2.0 statistical software.
Where: Y ij = the mean performance of the i th genotype in the j th environment, μ i = the grand mean of the i th genotype over all the environments,  i = the regression coeffi cient which measures the response of the i th genotype on environmental index, I j = the environmental index obtained by the difference between the mean of each environment and the grand mean and  ij = the deviation from regression of i th variety in the j th environment The pooled deviations mean square was tested against the pooled error mean square by the F-test to evaluate the signifi cance of the differences among the deviations of genotypes being evaluated from their expected performances.
As a result, in order to test the validity of the hypothesis that whether there is signifi cant difference among the 49 genotypes with respect to their mean grain yields or not and whether there is signifi cant difference among the regression coeffi cient or not, genotypes mean square and regression mean square were tested against the pooled deviation using the F-test.

Correlation and coeffi cient of determination
Spearman's correlation coeffi cient between different stability parameters and among agronomic traits and coeffi cient of determination (r 2 ) for grain yield of each genotype was estimated by using GenStat 18 th edition (2016) statistical software and Microsoft excel, respectively.

AMMI Stability Value (ASV)
In order to compute and rank genotypes according to their yield stability, the additive main effect and multiplicative interaction effect stability value (ASV) was proposed by Purchase [21]. It was calculated using Microsoft excel (2007) by employing the following formula: Where: ASV = AMMI'sstability value, IPCA1= interaction principal component analysis one, and IPCA 2= interaction principal component analysis II.
Similarly yield stability index (YSI) was also computed by summing up the ranks from ASV and mean grain yield [22]: Where: RASV is rank of AMMI stability value and RGY is rank of mean grain yield to statistically compare the stability analysis procedures used in the study.

Mean performance of genotypes
The overall performance of 49 sorghum genotypes tested based on mean grain yield and other agronomic traits across locations is presented in Tables 3. In this study days to fl owering, maturity, plant height, panicle length, grain yield and thousand grain weight were highly signifi cantly (P≤ 0.001)   affected by the combined effect of both genotype and growing conditions of locations, whereas days to emergence and grain fi lling period were non-signifi cant ( Table 4). The mean day to emergence at Humera was faster than the four locations.
This might be due to the amount and occurrence of rain fain and temperature at the time of plantation. The major environmental factors that affect germination of sorghum genotypes are temperature (including soil temperature), moisture and soil texture [23]. There was a variation among means of grain fi lling period of genotypes in the four locations.
The grand mean grain fi lling period of locations was 39 days, Humera and Kobo were the two locations that had faster grain fi lling period than the rest three locations. At Humera, genotypes fi lled their grains at a faster period than the genotypes in the other locations.

Grain yield
The mean grain yield obtained by the genotypes at the fi ve locations was 2184 kg ha -1 as shown in Table 4. The standard hybrid check ESH-1(G49) and K7252 (G29) produced higher mean grain yield with yield of 3278 and 3051 kg ha -1 respectively, whereas, G8 (5136) had the lowest mean grain yield with 735 kg ha -1 . However, the newly evaluated hybrids had not shown yield advantage over the standard hybrid check. In disagreement with this study, many researchers [15,[23][24][25][26][27] reported that tested varieties/ hybrids showed better performance than the best check for most of yield and other traits in sorghum.

Days to fl owering and maturity
Days to fl owering of the genotypes ranged between 58 to 69 days and the mean days to fl owering obtained was 63 days as shown in Table 4. The smallest days to fl owering was recorded by genotype 7270 (G39) while G14 (5160) had recorded shorter days to fl owering. The genotype K7445 (G7) had shorter days (95) to 90% maturity, whereas, the longest days to maturity (106) was recorded for genotype 5152 (G10).

Thousand grain weight
The average thousand grain weight (TGW) of the genotypes was 26.25g. The genotype with the maximum (28.3g) TGW was G19 (K7233) while genotype G4 (K7437) recoreded minimum (23.7g). Earliness is a very important trait under low-rainfall conditions. The trait having the most dominant effect on fi tting a plant to its environment for maximum productivity is the appropriate phenological development [30]. Conforming to the association among grain yield and other measured traits, the association between grain yield and days to fl owering was strongly negative (r = -0.580) and highly signifi cant (P≤ 0.001) while days to maturity was weakly negativly correlated with  The negative association between grain yield with days to fl owering and maturity indicated that moisture stress after fl owering might have caused a yield reduction in the late maturing genotypes, whereas, the early fl owering and early maturing genotypes had the advantage to fi lled grain early and escaped the moisture stress conditions. Similar results were reported by Kassahun, et al. [24], Taye, et al. [16] on sorghum;

Yield stability analysis
The following univariate stability analysis were performed for grain yield (kg/ha).

Eberhart and Russell's linear regression model
The analysis of variance by Eberhart and Russel's Model of striga resistant sorghum hybrids on mean grain yield (kg ha -1 ) tested at fi ve locations is presented in Table 6. Genotype  Coeffi cient of determination (r 2 ) represents the predictability of estimated response of the genotypes. The values of coeffi cient of determination ranged between 0.5662 for G14 and 0.9999 for G34, suggesting that linear regression accounted from 56.62% to 99.99%. This result showed that the variation in sorghum mean grain yield was explained by genotype response across the testing environments, which is in agreement with the previous fi ndings of Showemimo [33] in sorghum. Except one genotype (G14), all genotypes showed high coeffi cient of determination. However, seventeen out of 49 genotypes had yielded below average. Hence the interest of plant breeder is to develop genotypes with highest mean yield and which can be overcome by both predictable and unpredictable environment fl uctuations.

AMMI stability value
The result for stability analysis of genotypes using AMMI stability value (ASV) is given in Table 8. This stability analysis was based on the value of the fi rst two IPCA scores of genotypes.
According to this stability measure, the highest rank is given to the genotype that is close to the biplot origin, i.e, genotype that    Stability studies have allowed researchers to identify broadly adapted cultivar for use in breeding programs and have assisted to advance suggestions to farmers [34]. The most stable and adapted genotypes can be identifi ed using ASV as that of Lins and Binns method. Almeida, et al. [35], Vange, et al. [36], Abiy [23] and Zigale [37] also used this stability parameter to characterize the stability of sorghum.

Yield stability index
Genotypes with lowest estimated values of yield stability index (YSI) are desirable and considered as the most stable.
This result is in line with the fi ndings of Solomon, et al. [41] and Lalise [42] on maize. However, there was no signifi cant correlation between mean grain yield with Eberhart and Rusell's deviation from regression (S 2 di) (r=0.269) stability parameter and IPCA2 (r= -0.10).
The non-signifi cant correlation among yield and stability statistics indicated that, stability statistics provide information that cannot be collected from average yield alone. The high correlation among mean grain yield, bi, and r 2 is expected as the values of these statistics were higher for high yielding genotypes. The positive and signifi cant correlations between mean grain yield and r 2 , and bi and r 2 suggest that the parameter, r 2 should be considered only in measuring dimensions of grain yield, but could not adequately detect stability and, hence, its effi ciency in selecting desirable genotypes is limited when used alone. The same suggestion was given by Setegn and Habtu [43], Nigussie [44][45][46][47][48][49][50][51]. The negative correlation between grain yield and S 2 di indicated that high yielding genotypes may be associated with low S 2 di.    the high environmental variations and differential response of genotypes to the variable environments thus leading to inconsistent in ranking of genotypes. The mean grain yield of environments ranged from 588 kg ha -1 in E1 (Humera) to 4508 kg ha -1 in E5 (Sheraro).The highest yield was obtained from G49 (3278 kg ha -1 ), while the lowest was from G8 (735 kg ha -1 ).

Conclusion
The large sum of square and highly signifi cant environment effect indicated that the environments were diverse and caused most of the variation in grain yield. Therefore the largest proportion of the total variation in grain yield was attributed to environments. This indicates the existence of a considerable amount of differential response among the genotypes to the changes of growing environments and the differential discriminating ability of the test environments. interactions. Since the current study was conducted only for one year, the work should be repeated at least for some more years to give sound conclusions and reliable recommendations.