Genotype x environment interactions in Sorchum grown in the medium agricultural potential areas of Kenya
Fourteen genotypes, (four hybrids, six varieties,and four random mating populations) were grown in a randomized complete block design and three replications in five- locations for two seasons, i.e, 1979 long rains and 1979/80 short rains, in the medium agricultural potential areas of Kenya. Environment x Genotype interaction was studied for grajn yield/plot, grain yield/head, days to 50% flowering and mean plant height. The combined analysis of variance for each character indicated that genotype x location x season interaction (G x L x S) variance was highly significant and was also the most important component of genotype x environment (G x E) interaction. Both G x Land G x S interactions were not significant for any character except days to 50% flowering, although, the magnitude of G x S was higher than G x L interaction component in all cases. The nature of G x E interactions were investigated by means of regression analysis techniques of Finlay and Wilkinson (1963), Eberhart and Russell (1966) and Perkins and Jinks Cl968a and b l, I ( v ) It was concluded th~t a considerable portion of G x E interaction sum of squares (SS) could be attributed to the linear regressions in case of grain yield/plot (67%), grain yield/head (70%), plant height (83%) and days to 50% flowering (45%). For all characters, the pooled deviations ·from regressions were highly significant. In case of grain yield/plot and days to 50% flowering, G x E (linear) SS was not significant, showing more environmental effect on genotypes to express themselves in different environments. For grain yield/head and mean plant height, this variance was significant, showing that there are genetic differences among the genotypes for their regression on the environmental index. From the joint regression analysis, it was shown that for grain yield/ plot, the predictions of GE interaction based on linear regression are difficult to make. While for grain yield/head and days to 50% flowering, reliable predictions can be made only for some genotypes. However, such predictions in case of plant height were more practical. A number of adaptability and stability parameters were estimated. Adaptability was referred to the response of a particular genotype to environments and was determined by means of regression coefficient value (bi). Genotypes with bi = 1.00 were widely adapted; those with bi < L were adapted specifically to the unfavourable environments and those with bi >1 were adapted specifically to the favourable environments. The stability was referred to as the ability to show minimum interaction with the environments. The stability parameters taken into consideration were 'Phenotypic stability factor' (PS), 'Ecovalence' (Wi), 'Co efficient 0 f de term ination' (R .2), and 1 'deviations from the regression' (S~). High correlation was found between the ranks of genotypes according to Wi, Ri 2 and S~ parameters. The defects of PS parameter were pointed out. Hence, the stability of the genotypes was based on the value of S~ alone. The genotypes with the lowest S~ being the most stable and vice versa. The hybrids gave the highest t yield and were early at each environment, although they were slightly taller than the varieties. The hybrids, however, lacked stability and in general were more specific in adaptation. The populations tended to be taller, and their adaptability and stability were better. (vii) The difference in adaptability and stability of the different genotype groups was attributed to their different buffering mechanisms. It was suggested that the adaptability and stability of populations be fully exploited to realize their full potential. Considering the overall performance, the hybrids were the most desirable genotypes. HYBRID, which had wide adaptability, was most desirable together with HIJACK, which performed better in unfavorable environments and HIMIDI which performed better in the favourable environments. Among the varieties, MY 57 and 50 x 135/13/3/1, and among the populations, SERERE ELITE an& RS/R appeared most promising. A comparison of standard error of genotype means suggested that four locations, four seasons and three replications were optimum for such studies. The low intraclass correlations for all characters indicated that the location and seasons could be treated as random environments. Hence, seasons and locations can be used interchangeably when considering the allocation of plots for evaluating the genotypes.