Direct Variance-Covariance Modeling as an Alternative to the Traditional Guide Curve Approach for Prediction of Dominant Heights

A new method for developing dominant height prediction models, which is closely related to certain earlier guide curve-based methods, is derived from linear prediction theory by directly modeling variances and covariances of height remeasurements without involving any local parameters. This method relies on population-averaged mean height-age curve and covariances (correlations and variances/standard deviations) between heights at different ages for height projection given prior observations for a new subject. We constructed a new height model consisting of the means, correlations, and variances approximated, respectively, by the Chapman-Richards model, first-order continuous autoregressive model, and the power of the mean function. The new model was compared with four difference models derived from the (generalized) algebraic difference approach, by evaluating their projection accuracy using a second-rotation loblolly pine (Pinus taeda L.) data set collected from a well-designed experiment. Both kinds of models produced very comparable predictions, but the new guide model can be preferred for height projection using earlier (e.g., before age 10) measurements.