Novel methods for improved tree breeding

The development and implementation of statistical tools to improve inference in sustainable forest tree breeding are presented here. By combining classical quantitative genetic theory and novel statistical methods, a number of parameters are optimized. The results obtained are compared to those achieved by traditional methods for visualizing improvements to genetic parameters. The methods are tested on both simulated data and on a real Scots pine pedigree. Modeling non-additive gene action using a finite loci model indicates that the development of the additive variance component does not decay initially as the underlying theory predicts. This phenomenon is shown for different sets of genetic components and models. In addition, variable numbers of loci were used so that different numbers of interactions could be captured. To draw inferences about the genetic parameters, a powerful Bayesian Markov chain Monte Carlo method was developed. The method utilizes transformation of the genetic covariance structure to improve computational speed. By combining two different Bayesian Gibbs samplers, a useful hybrid sampler was developed; this was found to enhance convergence statistics and computational speed. A method that finds the number of trees and their respective mating proportions that will maximize genetic gain was implemented and modified to handle a large number of selection candidates. When testing the selection method on a real pedigree an increase in genetic gain of up to 30 % was found compared to traditional methods in which the same restrictions were placed on relatedness. In order to provide a long-term breeding perspective, the selection method was combined with various mating schemes to examine the development of genetic parameters. A modified minimum coancestry mating scheme resulted in a level of genetic gain closest to the theoretically achievable limit while reducing the level of inbreeding in the population.