Pattern analysis of genotype x environment interactions and comparisons with alternative analyses : a thesis presented in partial fulfilment of the requirements for the degree of Master of Science in Plant Science at Massey University

Abstract

The occurrence of genotype-environment interactions is a problem affecting the interpretation of cultivar trials. Several analyses have been used to try to resolve the inconsistencies of cultivar performances which occur when these interactions are present. An assessment of several techniques was carried out using three sets of data. Two sets of barley data came from one. season's trials covering the barley growing areas of New Zealand. Ten wheat cultivars were tested in four locations in the lower North Island in two seasons. The analyses which were examined were Analysis of variance, linear regression, Cluster Analysis and principal Component Analysis. The parameters of Wricke, Hanson and Eberhart and Russell were also studied. The Analysis of Variance revealed significant location, genotype and genotype-location interaction effects for the barley data. The wheat data had significant years, years x locations, genotype and genotype-year interactions effects. There was a strong linear relationship between the genotype means and the environmental index for the Finlay-Wilkinson regression analysis. Following refinement of the error term βi's with significant differences from 1.0 could be seen for several Barley and Wheat genotypes. It was noted that a

conflict existed between the aim of finding significant differences from 1.0 and the assumption of independence of effects for the underlying model. It was suggested that an independent measure of environments be used. The parameters of Wricke, Hanson and Eberhart & Russell were each related to different concepts of stability and the genotypes ranked accordingly. The three parameters gave reasonably consistent results for the rankings of the cultivars. In the barley data the cultivars Goldmarker and Magnum had uniformly low rankings. The wheat cultivar Gamenya was generally found to rank highly. These were measures of variability over environments so a high ranking infers a low level of variability and vice versa. A comparison of the different clustering strategies available was carried out and Wards Incremental Sums of Squares method was chosen as the major strategy. This was applied to each data set using both genotype-environment effects and means. A probabilistic cut off measure was used for truncation of the dendrogram. The clusters formed could be related to the previous analyses and seemed to adequately summarise the different responses present. A Principal Component Analysis was carried out and the number of components needed to account for 75% of the total variation were examined. For the barley data sets relatively large numbers of components were needed for this ( five and six). This made interpretation and presentation of the genotypic performances difficult. For the wheat data two components explained a satisfactory level of the total variation and the arrangement of the genotypes on these two axes agreed closely with the clustering results. Varimax rotation did not aid greatly in the interpretation of the components. It was felt that the roles of these different analyses were complementary in interpreting genotypic performances.