Generalized additive models reveal among-stand variation in live tree biomass equations

Abstract

Accurate estimation of forest biomass is important for scientists and policymakers interested in carbon accounting, nutrient cycling, and forest resilience. Estimates often rely on the allometry of trees; however, limited datasets, uncertainty in model form, and unaccounted-for sources of variation warrant a reexamination of allometric relationships using modern statistical techniques. We asked the following questions: (i) Is there among-stand variation in allometric relationships? (ii) Is there nonlinearity in allometric relationships? (iii) Can among-stand variation or nonlinearities in allometric equations be attributed to differences in stand age? (iv) What are the implications for biomass estimation? To answer these questions, we synthesized a dataset of small trees from six different studies in the White Mountains of New Hampshire. We compared the performance of generalized additive models (GAMs) and linear models and found that GAMs consistently outperform linear models. The best-fitting model indicates that allometries vary among both stands and species and contain subtle nonlinearities that are themselves variable by species. Using a planned contrasts analysis, we were able to attribute some of the observed among-stand heterogeneity to differences in stand age. However, variability in these results points to additional sources of stand-level heterogeneity, which if identified could improve the accuracy of live tree biomass estimation.