An evaluation of cultivar stability in ARC maize trials over a six year period

English: 1. This study was undertaken to compare various statistical methods of analysis to determine the most suitable procedure to evaluate maize genotype performance under the variable production conditions in South Africa, as well as to assess the suitability of these statistical procedures for characterizing yield stability. The main objective of this study was to recommend the most appropriate statistical procedure(s) to estimate maize genotype performance and stability more accurately and to investigate the GxE interaction and stability performance of genotypes in various environments by applying different statistical methods of analysis in order to make useful recommendations for future utilization. 2. Ninety four genotypes were planted at 80 localities in South Africa for the period 1998 to 2003. These trails, which were done by the Agricultural Research Counsel, were divided into six research regions. Three of these regions were used for the purpose of this study. Twenty four cultivars were planted at three sites under irrigation for the Irrigation research region, 24 cultivars at six sites for the Eastern region and 21 genotypes at five sites for the Western region for the period of 2001 to 2003. Both the Eastern and Western region are dryland maize production regions. Grain yield was determined and genotypes were evaluated for performance and yield stability in all three regions according to eight statistical procedures, which were (i) Combined Analysis of Variance, (ii) Cultivar Superiority Measure (Pi), (iii) Wi-Ecovalence (Wi), (iv) Shukla’s procedure of stability variance, (v) Stability Variance with Locality as Covariate, (vi) Rank differences (S1) and Variance differences (S2), (vii) Eberhart and Russell Regression Model, (viii) Additive Main Effects and Multiplicative Interaction (AMMI). The procedure proposed by Purchase (1997) to determine the absolute stability measure for the AMMI model was used as well. Comparisons between results from all statistical procedures as well as recommendations from Phurcase (1997) were used to determine the best suited procedure. 3. The combined analysis of variance ranked cultivars according to their mean grain yield measured in ton ha-1. CRN3505, CRN3549, PAN6844, Phb30H22 and CRN3760 indicated good mean yield over all three different regions. On the other hand QS7608, SC405, LS8508 and SC401 delivered the lowest mean yield values over all three regions. This procedure is no indication of the stability for yield of the genotypes. 4. Lin and Binns cultivar superiority measure indicated good yield stability, according to their definition and procedure, for especially the genotypes CRN3505, CRN3549, SNK2472, PAN6146. The worst performers were PAN6777, QS7608 and SC401. There was considerable correspondence in the performance of the genotypes over the Irrigation and Eastern regions as well as the Eastern and Western (dryland condition) regions. 5. Wricke’s ecovalence, Shukla’s stability variance and stability variance with locality as covariate procedures had approximately the same results. They showed different genotypes to be more stable in different regions. These three procedures found SNK2472, LS8508, SNK2778 and Phb30G03 to be the most stable cultivars and Phb30H22 and PAN6777 to the most unstable cultivars in the Irrigation region. For the eastern region PAN6757, CRN3549, SNK2778 and SNK2969 were the most stable genotypes while PAN6777 and SC405 were the unstable genotypes. Phb3442, SNK2472, PAN6615 and PAN6043 were the most stable cultivars in the western region with PAN6734 and SC401 the most unstable ones. 6. Rank difference and variance difference procedure correlated the best with the AMMI model. This procedure found SNK2472, LS8508, SNK2778 and NS2900 as the most stable genotypes in the irrigation region, while QS7608 and CRN3505 were the unstable genotypes. These results were very much the same as the previous three procedure’s results. But for the eastern region the results were different for the previous three procedures. PAN6777, SNK2472, PAN6757 and CRN3549 were the most stable cultivars and Phb30H22 and LS8508 were the unstable genotypes. In the western region, PAN6615, PAN6043, QS7608 and Phb30H22 were the most stable genotypes while SNK2900 and SC401 were the most unstable genotypes. 7. The Eberhart and Russel procedure, based on deviation from the regression in regression analysis, showed exactly the same results as Stability variance with locality as covariate and therefore had exactly the same rankings for all three regions as Wricke’s ecovalence, Shukla’s stability variance and stability variance with locality as covariate. 8. For the AMMI method, a procedure combining the IPCA1 and IPCA2 scores was used to determine an absolute AMMI value. According to this analysis, Phb30H22, PAN6740, QS7608 and CRN3604 were the most stable genotypes in the irrigation region. This differs totally from the results found by the other procedures. PAN6568 and CRN3505 were the most unstable genotypes. For the eastern region SC405, LS8508, QS7608 and SNK2472 were the most stable cultivars and Phb30H22 and PAN6777 were the most unstable cultivars. In the western region PAN6734, PAN6146, Phb30H22 and Phb3442 were the most stable genotypes and PAN6043 and PAN6479 were the most unstable genotypes. 9. Wricke’s ecovalence, Shukla’s stability analysis and Eberhart and Russel’s procedures correlated highly significantly. However, no correlation was found in the pair wise comparison of Lin and Binns’ (Cultivar Superiority Measure) procedure with the other procedures and very little correlation was found in the pair wise comparison of the AMMI model and the other procedures. The more holistic approach of AMMI is particularly effective in clarifying GxE interactions. Using the IPCA1 and IPCA2 scores to determine an AMMI stability value, superiority ranking of genotypes can easily be done. The AMMI model can summarize patterns and relationships of genotypes and environments successfully and offers a valuable prediction assessment. For this reason, it is recommended that this model is used for analyzing GxE interaction of maize genotypes in South Africa. The Lin and Binns procedure appears to be more of a cultivar performance measure than a stability measure which explains its correlation with the mean yield of the genotypes. The Eberhart and Russel, Wricke’s ecovalence and Shukla’s stability analysis showed highly significant correlation and are recommended to be used to describe genotype stability of maize genotypes rather than to be used to describe GxE interactions, due to the limitations of these techniques. 10. There were six genotypes and three localities that were common over the six year period for the irrigation region and six genotypes with six localities for the eastern region and seven genotypes with five localities for the western region. Wricke’s stability parameter was used to determine the most stable cultivars over all six years over all localities for all three regions. SNK2778 was the most stable genotype for the irrigation maize production region in South Africa for a six year period. In the eastern region, Phb3442 was the most stable cultivar with the lowest Wi value, while NS9100 was the most stable cultivar in the western region. The genotypes planted in the western region are different from those planted in the eastern and irrigation regions. This is due to the difference in climatic conditions. The western region is much dryer and warmer and therefore needs genotypes that are more stable under these extreme conditions. PAN6479 had one of the lowest mean yields throughout all three regions and was found to be relatively unstable in all the analyses. CRN3760 had a good mean yield and intermediate stability throughout all three regions over the period 1998-2003. 11. In the yield progress study, no significant progress was visible. Eighty cultivars planted for a six year period at Cradock were used for this analysis. Year 1998 had the lowest mean yield and 2000 the highest mean yield. The cultivars that were tested in the later part of the period 1998-2003, had overall higher mean yield, which proves that progress has been made by breeding companies in genotype yield increase.