A case for unreplicated plots for multi-site yield testing in wheat

Abstract

The effectiveness of unreplicated multi-site yield tests was evaluated by using data from 82 trials involving 16 sets of entries including cultivars or advanced breeding lines each grown on four-seven sites per year over an 11-year period. For each data set, entry means and ranks over sites were calculated using, first, all replications (usually four, but sometimes three) at each site; and, second, one of the four (or three) replications selected at random at each site. The estimates of entry mean yields over sites by using only one replicate were highly and positively correlated with estimates using all available data. The correlations varied from +0.88 to +0.98 with a mean, over all 16 data sets, of +0.94. In addition, if the 10 highest yielding entries were selected on the basis of unreplicated trials at several sites, the top five entries according to multi-replicate trials were invariably retained. It is concluded that unreplicated multi-site tests offer breeders an effective and efficient means of identifying high yielding genotypes, particularly early in the breeding program when seed supplies are limited.