Regeneration imputation models for the ICHmw2 subzone in the vicinity of Nelson, BC

Forests are dynamic systems, seldom in equilibrium. This is due, in part, to anthropogenic disturbances such as harvesting. Understanding the dynamics of complex stands has become a management priority and is the subject of a number of studies in northwestern North America. Understanding regeneration patterns in these stands is crucial, since future stands are determined by the way the regeneration is managed. The objective of this study was to explore and test the applicability of using imputation techniques rather than more traditional regression techniques for predicting regeneration in the complex mixed-species stands prevalent in the Interior Cedar-Hemlock moist warm subzone variant 2 (ICHmw2) in the vicinity of Nelson, BC. Two approaches were used: tabular imputation and most similar neighbour (MSN) imputation. The regeneration data collected during the 2000 field season and that collected during the 1998 field season are summarized. A series of tabular imputation and MSN imputation approaches are developed and their performances in predicting the regeneration are compared. The tabular approach depicted average regeneration by five site groups, two residual density classes, five time-since-disturbance classes, species, and height classes. The MSN approach made use of regeneration data of some plots (called reference plots) and a complete coverage of selected easy-to measure attributes for the entire data set (called reference and target plots) for its development. MSN imputation provided regeneration for the assumed missing regeneration data (target plots) by choosing a most similar plot from the reference plots to act as its surrogate. The most similar plot selection was based on a similarity measure that took into consideration the multivariate relationships between the two different sets of data. The full MSN imputation model (four height classes) was the best predictor for regeneration. Stand density indicators (basal area, number of residual trees per hectare, and crown competition factor) were the driving variables in the most similar neighbour selection process. When the number of match categories and the root mean square error (RMSE) were used as comparison criteria, about 97.5% of the target plots were classified as being moderate to good. Perfect matches with high precision corresponded to those plots that had high number of cells with no regeneration (zero). The mismatch of basal area, trees per ha, crown competition factor, and seemingly, the presence/absence of advance regeneration seemed to be a major cause of poor predictions. A sensitivity analysis showed that Prognosis[sup BC] was mostly insensitive to regeneration predictions from both imputation models during the first 50 years of the projection. However, regeneration estimates were generally good. Also, with longer periods of simulation, it is likely that the model would be more sensitive, particularly to tabular predictions. As Prognosis[sup BC] grows stands based on the interaction among trees, a user can provide data either by using the selected most similar neighbour plot, or by using the means from the table that has the desired characteristics. As more data become available, these tables can be easily updated and the reliability of tables based on small sample size can be improved. There were not a lot of obvious trends apparent in the tabulated data. This may be due to the dominance of advance regeneration among the regeneration present. Advance regeneration would be more affected by the conditions that existed prior to the most recent disturbance than the conditions that exist today. Designing a sampling method that separates advance from subsequent regeneration will, without doubt, improve the results.Forestry, Faculty ofGraduat