MATERIALS AND METHODS

Fifteen oil palm genotypes from the Nigerian Institute for Oil Palm Research (NIFOR) Main Breeding Programme were evaluated over a period of four years (1999-2002) in a tropical rainforest zone of Benin City, Edo State, Nigeria (6° 31’N and 5° 40’E). The experimental layout was a Randomized Complete Block Design (RCBD) with six replications. A spacing of 9 meters triangular was adopted while fertilizer application and other cultural practices were as recommended by NIFOR. Fresh Fruit Bunch (FFB) yield (kg palm^-1) was recorded in-situ in the field for four years for each progeny.

Progeny means were subjected to pooled analysis of variance. Genotypes were assumed to be fixed while year and replicate effects were random. Stability analysis was performed using the joint linear regression method of (Eberhart and Russell, 1966) (ER), the stability index of Shukla (1972) (SH), the Coefficient of Variation (CV) of Francis and Kannenberg (1978) (FK), the cultivar performance measure (P_i ) of Lin and Binns (1988) (LB) and the relative adaptation technique in the GGE biplot of Yan (1999) and Yan et al. (2000) (GGE). The GGE biplot was facilitated by the GGE software (Yan, 2001).

Stability of genotypes were defined as regression coefficient of b_i = 1 and deviation from the regression as small as possible (S²d_i = 0) for ER method, high mean yield and consistent low CV for FK method, low stability variance for SH method, low cultivar performance for LB method and a lower absolute value for GGE method.

To statistically compare the above stability procedures, Spearman’s coefficient of rank correlation (r_s) was determined (Steel and Torrie, 1980). All the genotypes were ranked according to the assigned values for each procedure’s analyses and definitions. Concordance analysis was carried out to determine the proportion of genotypes that were jointly selected as stable by two or more analytical methods. The concordance coefficient (CC_ij) for any two methods was calculated as follows:

Where:

A concordance coefficient approaching zero would indicate complete divergence of analytical methods while a coefficient near unity would suggest a high level of convergence of the methods.

RESULTS AND DISCUSSION

The analysis of variance table for the 15 oil palm genotypes showed significant differences among genotypes, environment and genotype x environment interaction (Table 1). The significant gxe suggests wide adaptation of the genotypes. The large proportion of environmental variance indicates the enormous influence of environment on yield performance of oil palm hybrids in Nigeria. This result is in tandem with the earlier reports of Asfaw et al. (2008) in small red beans and Jalata (2011) in barely.

According to ER method, it is specifically the deviation from the regression (S²di) which is used as a measure of genotype’s stability across environments. The most stable genotypes with the lowest S²d_i values were DT 9, DT 7, DT 1, DT 11, DT 6, DT 13 and DT 15 (Table 2). The most unstable genotypes with the highest S²d_i values were DT 12, DT 5, DT 3, DT 10 and DT 8.

If the mean bunch yield, regression coefficient value and the deviation from the regression (S²d_i) are considered together, then the most widely adapted would be DT 7 and DT 9 because their regression coefficients were close to unity (b_i = 1), S²d_i = 0 and high bunch yield (Fig. 1). The genotype, DT 10 had high bunch yield and the regression coefficient was nearly equal to one. It was however considered unstable because S²d_i >1 (Table 2, Fig. 1). These present findings were in agreement with earlier investigations of Sreedhar et al. (2011) on yield stability in hybrid rice.

Table 1:	Analysis of variance of fresh fruit bunch yield of 15 oil palm genotypes grown in 4 environments

Table 2:	Fresh Fruit Bunch (FFB) yield and associated stability parameters of 15 oil palm genotypes grown in four environments
S2di is represented in column 3 as ER

Fig. 1:	Regression coefficient plotted against mean fresh fruit bunch yield (kg/palm/year)

The most stable genotype as indicated by SH method were DT 7, DT 9, DT 11 and DT 15 (Table 2). The genotypes with poor stability according to this parameter were DT 5, DT 10, DT 12, DT 3 and DT 1.

Following the genotype grouping technique, five genotypes using the FK method were considered to be highly stable (Fig. 2, Table 2) because of the high mean bunch yield and low CV. The group II genotypes produced bunch yield above the grand mean but had high CV which made them unreliable. Yield consistency is more important than absolute yield to subsistence farmers who need to be assured of harvest from season to season or year to year (Evans, 1993; Hassanpanah, 2010). The genotypes in group III had low CV but produced below grand mean bunch yield. They are therefore, considered stable and could be selected for further improvement by increasing their yield potential. FK method will be amenable to use especially in screening a large number of genotypes for yield stability. This conforms to Ataga (2010) report in the yield stability study of oil palm using descriptive method of grouping genotypes.

Using the cultivar performance measure (LB method), the genotypes with the lowest p_i values (DT 7, DT 11, DT 15, DT 9 and DT 8) were judged most stable (Table 2).

The most unstable genotypes according to this analysis were DT 5, DT 13, DT 1, DT 4 and DT 10. The genotypes DT 6, DT 2 and DT 3 which respectively ranked 3rd to 5th for mean FFB yield, showed intermediate stability and ranked 6th to 8th for LB.

The visual display of the average yield and stability of genotypes from the GGE biplot method classified DT 15, DT 9, DT 13, DT 7 and DT 11 as the most stable genotypes for FFB yield due to the lesser absolute values (Fig. 3, Table 2). It is however worthy to note that genotype DT 13 has the lowest bunch yield performance. This model selected DT 12, DT 3, DT 2, DT 5 and DT 14 as the least stable genotypes because of the greater absolute values.

The Spearman’s rank correlation among environments and stability indices of the five methods based on ranking of FFB yield of 15 oil palm genotypes was generally weak (Table 3).

According to Huhn (1996), this is an indication of cross order pattern due to strong gxe interaction. Consequently, the order of ranking of genotypes in an environment would not predict the pattern in other environments.

Fig. 2:	Mean FFB yield (kg/palm/year) plotted against CV (%)

Fig. 3:	Average tester coordinate (ATC) view of FFB yield for 15 oil palm genotypes

Table 3:	Rank correlations among environments and stability parameters of five stability methods for 15 oil palm genotypes
** Correlation is significant @ 0.01 level, * Correlation is significant @ 0.05 level

The bunch yield at E99 had the highest correlation (r = 0.854, p<0.01) with the mean FFB yield over all the environments. It could be inferred that the mean performance at E99 may be representative of the performance at the other three environments. Shukla’s stability index procedure (SH) was highly significantly correlated with ER deviation parameter and LB cultivar performance value. Martin (2004) reported highly significant correspondence between SH and ER procedures in his comparative studies of statistical methods to describe gxe interactions and yield stability in multi-location maize trials. This may suggest similarity to the procedures of ER and LB. GGE biplot method showed the greatest deviation from all other procedures, having negative rank correlation coefficients compared to the other procedures (Table 3). The weak correlation among the stability procedures may be due to genotypes sensitivity to different procedures as a result of the principles underlying the choice of stable genotypes.

There was a high concordance among the stability models (Table 4). The level of convergence between any two of the stability models ranged between 25 and 67%. The highest agreement (66.7%) on stable genotypes was between ER and SH, SH and LB, SH and GGE and LB and GGE. These were followed by ER and LB, ER and GGE, FK and SH, FK and LB and FK and GGE (42.9%).

Table 4:	Concordance analysis for number of genotypes jointly selected by two or more analytical methods
aMethod 1 = ER, Method 2 = FK, Method 3 = SH, Method 4 = LB, Method 5 = GGE, bCC = Concordance

However, the concordance between any of the three or five methods was between 28.6 and 57.1%. The high percent divergence in the choice of stable genotypes as seen in this study is similar to earlier reports of Lin et al. (1986), Becker and Leon (1988) and Baiyeri et al. (1999). It is interesting to note that there were genotypes selected as stable by the five methods based on a 28.6% concordance (Table 4). These genotypes are superior and therefore more reliable for future breeding programmes (Papadopoulos et al., 2007).

Although, there may be some differences in the identification and selection of stable genotypes using different stability procedures, stability models with the same concordance percentage is an indication of similarity in selection efficiency of stable genotypes. This suggests availability of close substitutes or alternative procedures. For instance, the high level of concordance between methods 1.5, 2.5, 3.5 and 4.5 shows that each of the methods could be used in lieu of method 5 (GGE) whilst achieving the same results. This is especially when the model is very complex, rare or expensive for the scientist or researcher.

CONCLUSION

Fresh fruit bunch yield showed significant differences for the genotype, environment and gxe interaction terms. The marked influence of environment on the magnitude of genetic variance suggests the responsiveness of FFB yield to environmental conditions or fluctuations. The combined use of the five stability procedures implicated genotypes DT 7 and DT 11 to be highly stable for FFB yield. These genotypes will be very reliable for future breeding programmes to develop high yielding planting materials with stable performance. Notably however, DT 5 which produced high FFB yield was rated as unstable by all the stability methods. Hence, this study shows that high yielding genotypes may not necessarily be highly stable over environments. The high percent divergence in the number of genotypes identified as stable is due to the inherent variation associated with the respective procedures for identification of stable genotypes. It has been concluded that the simultaneous use of different statistics would protect the breeders from grievous errors of selecting a wrong genotype especially when testing new varieties. In addition, the high concordance observed in some of the methods suggests availability of alternative procedures. It is therefore recommended that ER, FK, SH and LB methods could be used in lieu of the GGE biplot method particularly where the software is not available.

ACKNOWLEDGMENT

The authors are very grateful to Mr. Sam Ofodile for several thought provoking questions about stability parameters and patterns, Mrs. M. T. Okoye for her great help in revising this manuscript and the executive director of NIFOR for permission to publish this study.