Improving precision of breeding values by removing spatially autocorrelated variation in forestry field experiments

Summary Adjustment for micro-environmental heterogeneity in inadequately blocked field experiments is desirable to improve estimates of genetic parameters and to maximize genetic gains from selection. In three forestry field trials of red alder (Alnus rubra Bong.) we removed spatially autocorrelated variation with kriging and evaluated the effect on estimates of treatment means and heritabilitiy compared to standard analysis. Kriging removed block effects and reduced the family × block interaction in all traits. The variation due to interactions was recovered in simple family or provenance variance components, which increased by up to 40 % in some traits. Heritability estimates and expected gains from selection increased accordingly, while the standard error of the estimate for family and provenance means decreased. The improvement was largest in experiments where blocking was clearly inadequate to capture site variation, when block size was large, and for traits that could be influenced by variation in soil properties. Bud break and leaf abscission, which are presumed to be independent from variation in soil, were spatially independent. Heritabilities estimated from an experiment with incomplete block design with nine trees per block could only be minimally improved. We recommend that variograms should routinely be constructed in the analysis of forestry field trials to test if residuals from standard models are spatially autocorrelated. If they are, kriging is proposed as a useful supplement to ANOVA in tree breeding experiments and other forestry field trials.