Model-dependent sampling for timber value in old-growth forests of coastal British Columbia

The procedure used to sample crown timber before harvesting in B.C. is designed to estimate net volume per ha using systematically located angle-count plots where trees are selected with probability proportional to basal area. The primary purpose of the sample is to provide information for timber valuation and stumpage appraisal. Timber value is the most important population parameter for stumpage calculation, but it is not explicitly considered in the sampling design. The objective of this study was to modify the current sampling method to increase the efficiency for estimating value using model-dependent sampling theory. Eighteen model-dependent sampling strategies were developed from six subsampling methods using three estimators. The six subsampling methods were used to select trees from angle-count plots to estimate the relationship between cruiser-called and estimated tree value. Three subsampling methods used probability-based selection of trees and three methods used purposive-based selection of trees. Ratio, average ratio, and regression estimators were used with each method. The 18 strategies were tested using Monte Carlo simulation with 2000 samples at each of nine sample sizes in three test populations. The test populations were created by grouping angle-count plot data into mutually exclusive sets reflecting different stand characteristics. The sample sizes were n = 20,40, and 60 plots with m = n, 3n, and 5n subsampled trees. Individual tree value was estimated with regression equations that used variables closely related to the value of each species. The sampling strategies were evaluated for bias, sample variance, achieved subsample size, sampling cost, confidence interval coverage, and relative advantage against the current sampling method. The model-dependent subsampling methods using purposive selection of trees were more efficient than the current sampling method considering cost and variance. The purposive-based methods were biased up to about 5%; the probability-based methods were slightly less biased. The two most efficient methods were: i) purposive selection of trees with the highest estimated values in a plot; and ii) purposive selection of trees with estimated values within a given range to give a second-stage sample balanced on the auxiliary variable. The greatest efficiency was always achieved with one sample tree per plot. The current sampling method was unbiased for estimating value but required approximately twice as many plots to estimate value to the same level of precision as net volume.Forestry, Faculty ofGraduat