Responsible reporting of forest inventory requires accounting for uncertainty both in tree measurements and in the allometric relationships used in biomass estimation. For example, IPCC guidelines require that country-level carbon accounting include estimates of uncertainties in forest biomass. It is rare, however, for all sources of error to be reported correctly. In the US Forest Service, following FIA (Forest Inventory and Analysis) protocols, field checks of production crews are routinely made by quality control crews, which provides a means to ensure that uncertainties are within acceptable bounds. The difference between the measurements made by these two crews is a rich source of data on measurement uncertainty that has yet to be reported to the scientific community. In addition to reporting the magnitude of measurement uncertainty in the FIA data from 24 states in the northeastern and north central USA, we also provide advice on propagating uncertainty in forest biomass models. We illustrate how to do this correctly by applying a random sample of the uncertainty in the model (the confidence interval) to all the trees simultaneously for each iteration, while applying the additional uncertainty in the prediction of individuals (the mean squared error of the model) randomly to each tree.
Results/Conclusions
Not surprisingly, tree diameter measurements (averaging 0.02 cm or 0.8% of tree diameter) have better agreement between the two crews than does actual tree height (1.1 m or 7% of tree height), with some species being more difficult to measure repeatedly than others. We also report errors in tree status (live, dead, or cut), crown class, crown ratio, grade, and decay class, and demonstrate how these affect confidence in estimates of forest biomass and timber value. The uncertainty in forest biomass estimates due to prediction of individuals using allometric regression models was large for plots with few trees; for plots with 30 trees or more, the uncertainty in individuals was less important than the uncertainty in the mean. Randomly applying prediction error to each tree results in absurdly small estimates of error in the case of very large numbers of trees; the uncertainty due to allometric models cannot be smaller than the model error itself. However, model error is often small compared to sampling error due to natural variability in forest inventory across large areas. Understanding the relative magnitude of all sources of uncertainty is important to making decisions about improving the design of forest monitoring systems.