Methods to evaluate wheat cultivar testing environments and improve cultivar selection protocols

Abstract

Analysis of cultivar by environment (C × E) interaction can improve efficiency of crop breeding efforts. Variety selection and recommendation based on wheat ( Triticum aestivum L.) yield testing trials could possibly benefit from this type of analysis as well. The objectives of the present work were to evaluate methods to identify relevant testing environments and improve the predictive value of data from wheat cultivar yield trials in the eastern US. The data come from 32 site years of winter wheat yield trials conducted in Virginia. Biplot analysis revealed that all current testing sites were relevant and that most performed similarly within a year. The degree of relationship or dissimilarity among environments was also evaluated using straight-line distance between observations in variable space measured as the squared Euclidean distance (ED). Analysis using the ED method revealed that all environments contained the centroid and were thus representative testing environments, similar to results from the biplot analysis. Biplots were effective at identifying cultivars and testing locations that were major sources of C × E interaction. Biplots and best linear unbiased predictions (BLUPs) were used to compare cultivar performance across environments. In a separate evaluation, the ED from the centroid to a cultivar mean was used to weight past relative yields for that cultivar and increased the predictability of future yield of a cultivar in three of four seasons. Weighting by ED decreased the number of site years needed to develop confidence in the yield stability of a particular cultivar from six to three. Utilizing the BLUPs for future grain yields, predictive ability of future performance after 1 year was 40% better and overall was 25% better than that achieved by weighting with ED. Overall the BLUPS method of estimating future performance was more accurate and more reliable than weighting with ED.