Modelling growth and yield of Douglas-fir using different interval lengths in the South Island of New Zealand
Degree GrantorUniversity of Canterbury
Degree NameDoctor of Philosophy
This study describes several refinements and improvements in whole stand growth and yield modelling of Douglas-fir grown in four regions of the South Island of New Zealand, namely Canterbury, Nelson, Southland and Westland. Modelling growth and forecasting yields are necessary for providing adequate tools with which to manage wood production from forests. The study comprised three major components: 1) development of whole stand growth and yield models with data sets of various interval lengths; 2) cross fitting models with different data sets reciprocally; and 3) check estimates using a growth and yield model derived from a data set free of auto-correlation. The methodology emphasised in developing the equations in this study involved rearrangement of the data to reflect different interval lengths among re-measurements for modelling purposes. Modification of data sets allowed an investigation into which growth intervals should be used to obtain the least biased models overall, and efficiently. The approach involved fitting single equations to each of three state variables, mean top height (h1OO), basal area/ha (0) and stocking/ha (N). Differences in growth trajectories across the four regions were identified and incorporated into single variable equations using dummy variables for improving the fitting of mean top height (h1OO), basal area/ha (0), and stocking/ha (N) equations. The main finding from this study was the level of improvement in making predictions through adoption of a mixed interval projection equation strategy compared with other options. Examining consistency among the predicting equations which had been developed from the different interval data sets, involved testing each form of model individually for all the data sets. The models based on mixed intervals were found to fit well for all the other interval length data sets. A subset of uncorrelated data was then created by selecting one re-measurement from each permanent sample plot (PSP), which was then used to validate the appropriateness of the equations derived from the full data sets, in order to overcome problems of dealing with correlated data. Coefficients for each of the equations for mean top height, basal area/ha and stocking/ha which were derived from this check data set were found to be very similar to the regression coefficients obtained from the full data set. Although the growth models developed in this study may require further examination, they do provide a very useful guide for selecting appropriate re-measurement interval lengths to derive satisfactory models which are the least biased overall. It is strongly recommended that modellers in the future adopt a mixed interval strategy as one data set option to evaluate.