Comparing effects of uncertainty in predictions of local and pantropical allometric models on large-area estimates for mean aboveground biomass per unit area

In the absence of regional/local allometric models of known accuracy, pantropical models (PMs) are often employed for predicting aboveground biomass (AGB) for trees growing in (sub)tropical forests. Using accurate models for a given population is crucial to increase accuracy and reduce uncertainty in estimates for mean AGB per unit area. This study evaluated the effects of local models (LMs) and PMs on large-area estimates for mean AGB (Mg ha$<^>{-1}$) in the Brazilian subtropical evergreen rainforest. In addition to the uncertainty due to sampling variability in the forest inventory dataset, uncertainty in model parameter estimates and residual variability were incorporated into standard errors (SEs) of the estimator of the mean through a Monte Carlo scheme. Generally, estimates for mean AGB were somewhat similar regardless of the model. Estimates for mean AGB obtained using a PM constructed with moist forest sites only and an LM were not statistically significantly different at significance level of 0.05. However, substantially less precise estimates for mean AGB were obtained with LMs constructed with 50 sample trees or fewer relative to an LM constructed with 105 trees and PMs, mainly as an indirect effect of greater uncertainty in model parameter estimates. When correlation among tree observations on the same sample location was accounted for when fitting the PMs, SEs increased as much as 26%. Further, although the PMs were constructed with many-fold larger datasets, they yielded less precise estimates for mean AGB than the LM constructed with 105 trees. Nevertheless, the evaluated PMs may still be regarded as accurate for the studied population.