Estimation of live tree biomass is an important task for both forest carbon accounting and studies of nutrient dynamics in forest ecosystems. In this study, we took advantage of an extensive felled-tree database (with 2885 foliage biomass observations) to compare different models and grouping schemes based on phylogenetic and geographic variation for predicting foliage biomass at the tree scale. We adopted a Bayesian hierarchical statistical framework, first to compare linear models that predict foliage biomass directly to models that separately estimate a foliage ratio as a component of total aboveground biomass, then to compare species specific models to both 'narrow' and 'broad' general biomass models using the best fitted functional form. We evaluated models by simulating new datasets from the posterior predictive distribution, using both summary statistics and visual assessments of model performance. Key findings of our study were: (1) simple linear models provided a better fit to our data than component ratio models, where total biomass and the foliar ratio are estimated separately; (2) species-specific equations provided the best predictive performance, and there was no advantage to narrow species groupings relative to broader groups; and (3) all three model schemes (i.e., species-specific models versus narrow or broad groupings proposed in national-scale biomass equations) tended to over-predict foliage biomass and resulted in predictions with very high uncertainty, particularly for large diameter trees. This analysis represents a fundamental shift in carbon accounting by employing felled-tree data to refine our understanding of uncertainty associated with component biomass estimates, and presents an ideal approach to account for tree-scale allometric model error when estimating forest carbon stocks. However, our results also highlight the need for substantial improvements to both available fitting data and models for foliage biomass before this approach is implemented within the context of greenhouse gas inventories.
Bibliographical noteFunding Information:
We wish to thank David Walker, Jereme Frank, Aaron Weiskittel, and all who have contributed to the U.S. Forest Service Volume Biomass Project and the Legacy database. In addition we would like to thank John Stanovick, David Bell, Kenneth Elgersma, and three anonymous reviewers for their comments on our manuscript. This research is funded by the U.S. Department of Agriculture, Forest Service, Northern Research Station and the Minnesota Agricultural Experiment Station .
© 2016 Elsevier B.V.
- Bayesian hierarchical models
- Component ratio models
- Foliage biomass models
- Posterior predictive checking
- Prediction uncertainty