Application and comparison of spatial models in analyzing tree-genetics field trials

Publisher Website
Google Scholar
Abstract
Tree height of jack pine full-sib families, originating from all possible combinations of three parental provenances and growing on three sites, was analyzed with 1 classical model and 11 nearest-neighbour spatial process models. Extension of the classical linear model with spatial interaction terms was deemed necessary in light of significant neighbourhood correlations among effect-free observations (residuals) on two of the three sites. The strength and extent of spatial and temporal correlations are demonstrated in both visual and tabular form. Only 4 of the 11 spatial models provided a substantial reduction (5–20%) in the significant difference between two estimates of full-sib family tree height. Spatial adjustments averaged 1–3% at the family level, with few families adjusted by more than 10%. The cumulative (temporal) effect of spatial covariance was demonstrated in rank correlations between adjusted and observed family means. No simple trends were obtained when adjusted variance components and heritabilities were compared with their unadjusted counter-parts, but most models tended to deflate genetic effects and reduce heritabilities. It is concluded that although spatial analyses provide an attractive tool for the experimenter, the lack of a cause and effect hypothesis in forest genetic trials necessitates model searching without the guarantee of true treatment effects. Spatial analysis provides good indicators of the need to collect additional site information for more powerful analyses. Careful planning and intensive site preparation may greatly reduce spatial covariances and the need for spatial analyses.