Accounting for error correlations in diameter increment modelling: a case study applied to northern hardwood stands in Quebec, Canada.

In this study, a diameter increment model was calibrated for individual trees in hardwood stands in Quebec, Canada. Two random effects, a first one for the plot and a second one for the time interval nested in the plot, and a covariance structure were included in the model to account for spatial and serial correlations. The diameter at breast height, species group, vigor and product classes, and basal area were the explanatory variables that were tested in this analysis. The adequacy of the covariance features (random effects and covariance structure) were tested through empirical correlations calculated from normalized residuals. A cross-validation was also carried out to evaluate the model. The normalized residuals showed no departure from the assumption of independently and normally distributed error terms with homogeneous variances. Consequently, the statistical inferences could be considered as valid. The results showed that the average diameter increment pattern differs among the species. Although tree product and vigor were significant explanatory variables, their effects were relatively small. On the other hand, basal area had a large and significant negative effect on diameter increment. Our study demonstrates that empirical correlations calculated from normalized residuals can be used as an additional tool to test the adequacy of the covariance features in a mixed-effects model. [ABSTRACT FROM AUTHOR]