NEAREST NEIGHBOR PROCEDURE AND DENSITY-DEPENDENT YIELD PREDICTION IN BARLEY (HORDEUM VULGARE L.)

Abstract

Agronomists are constantly experimenting with improved plot techniques that can enable them to make more precise inferences from field data. This dissertation reports two investigations: (a)evaluation of the yield potentials of some barley genotypes using two non-traditional methods, and (b)comparative assessment of the two methods. Two separate but related experiments were conducted. The nearest neighbor procedure was the first. The use of spaced-plant parameters to predict yield at normal commercial density was the second experiment. Four variations of the nearest neighbor procedure were examined. For each version the plant to be evaluated always occupied the center of the rectangle of nearest neighbors. Evaluation consisted of yield adjustments where the yield of the individual plot was compared with the mean of its nearest-neighbor genotypes. Individuals were ranked according to those deviations. Unadjusted yield data were also ranked. The error mean squares derived from ranks of various configurations were compared inter se and with that from unadjusted yield. Nearest neighbors always showed a smaller error variance than the unadjusted data. Of these the first nearest neighbors produced the smallest mean square for error and, hence, the highest efficiency of genotype ranking. This procedure substantially controlled for the effects of soil heterogeneity. Averages of individual ranks were computed and related to respective genotypes (entries). For each procedure the top 25% which fell in the upper bracket of the yield curve were considered to possess high yield potentials. This method of adjustment, ranking, averaging, and selection was applied to the unadjusted data as well as to each of the nearest neighbor procedures. Unadjusted mean yield and nearest neighbor techniques were contrasted. The rankings generated by the two procedures were similar but not identical. The significantly lower error variance of the nearest neighbor adjustments indicated that those should be used instead of unadjusted mean yield when precision is needed. However, unadjusted mean yield ranking provides broad identification of high yielding genotypes, and is a simpler statistical procedure. The second experiment examined the effectiveness of yield and yield components of spaced plants in predicting yield at normal cultural density. It was conducted for two years using primarily trend analysis. Results for individual years showed that none of the metric components of spaced plants was a satisfactory predictor of crop yield. However, when data were pooled over the whole experimental period, most of the yield components of spaced plants showed highly significant correlations with crop yield. Regression models were developed from the components which demonstrated good prediction of crop yield. Under the conditions of this study, productivity (biological yield or total weight) was revealed by all the analyses as the most important spaced-plant component for predicting yield at higher densities.