Analysis of early tree height in forest genetic trials is enhanced by including a spatially correlated residual

Abstract

This study explores the use of a mixed linear model including spatially correlated residuals in addition to the traditional randomized complete block (RCB) analysis in 12 forest genetic trials. The analysis of early height data from progeny and clonal tests of three species (Picea sitchensis (Bong.) Carr., Pinus pinaster Ait., and Pinus radiata D. Don) showed that there was significant spatial variation in all trials. Adding a basic first-order separable autoregressive error term more effectively modelled the spatial variation than the RCB model and greatly reduced the block and plot variances. There was no evidence that extended spatial modelling was required. The spatial analysis greatly improved the accuracy of genetic value estimation in some trials and was accompanied by large changes in rank of the genetic entries and by greater gains in selection relative to the RCB analysis.