USING BEST LINEAR UNBIASED PREDICTIONS TO ENHANCE BREEDING FOR YIELD IN SOYBEAN .2. SELECTION OF SUPERIOR CROSSES FROM A LIMITED NUMBER OF YIELD TRIALS

Economic constraints on many plant breeding programs have forced breed ers to Limit the number of environments for performance testing of new genetic material. The use of best linear unbiased predictions (BLUP), which augments predictions of individuals by using observations on th eir close relatives, should provide improved predictions of performanc e under such conditions. The objectives of this study were to determin e (i) whether BLUP values were more precise predictors than least squa res means [i.e, best linear unbiased estimates (BLUE)] from soybean [G lycine max (L.) Merr.] yield trials conducted in one or two environmen ts, and (ii) how much improvement in the precision of BLUP could be ga ined by inclusion of historical parental data. Bulks and lines of 24 s oybean crosses and four check cultivars were evaluated at 11 different environments in Tennessee to estimate the mean seed yield of each cro ss and cultivar. Historical yield records on parents of Each cross wer e compiled from trials conducted in Tennessee from 1982 through 1990. Using subsets of the 11 environments, we predicted yield using three m ethods: (i) BLUE, (ii) BLUP(NP), without parental data, and (iii) BLUP (P), with parental data. Standard errors of differences (s(d)(-)) and rank correlations (r(s)) between the actual and predicted mean yields showed that either method of BLUB was superior to BLUE for providing p recise yield estimates. Because of the high genetic relationships amon g the crosses used in this study, including historical parental inform ation did little to increase the precision of BLUP(P) over BLUP(NP).