A new composite k-tree estimator of stem density

This study presents a generally applicable and robust k-tree composite estimator of density. We propose to estimate stem density by a weighted average $ \left( {\hat{\lambda }_{\text{aic}} } \right) $ of 16 individual density estimators. The weights given to individual estimators are inversely proportional to the relative fit (Akaike’s corrected information criterion) of each estimator to the assumed distribution of observed k-tree distances. The performance of the proposed estimator is evaluated in simulated simple random sampling with k?=?3 and 6 in 58 forest stands (54 actual and 4 simulated) and 600 replications. Sample sizes were 15 and 30 locations per stand. Eleven estimators were novel, including three designed for regular spatial patterns. Absolute stand-level bias with k?=?6 varied from 0.1 to 8.1% (mean 1.8%), and a bias larger than 6% was limited to 3 stands with either pronounced density gradients or a strong clustering of stem locations. Root mean squared errors were approximately 16% (k?=?6 and n?=?15) versus 12% for sampling with comparable fixed-area plots. Coverage of computed 95% confidence intervals ranged from 0.72 to 0.99 (median?=?0.98 with n?=?15 and 0.95 with n?=?30), with 98% of all intervals achieving a coverage of 0.85 or better. In seven stands used in an assessment of a novel spatial point pattern reconstruction k-tree density estimator (RDE) by Nothdurft et al. (Can J For Res 40:953–967, 2010), the average absolute bias of $ \hat{\lambda }_{\text{aic}} $ with k?=?6 was 1.5 versus 0.7% for $ \hat{\lambda }_{\text{RDE}} $ .