Evaluation of alternative approaches for landscape-scale biomass estimation in a mixed-species northern forest

There is growing interest in estimating and mapping biomass and carbon content of forests across large landscapes. LiDAR-based inventory methods are increasingly common and have been successfully implemented in multiple forest types. Asner et al. (2011) developed a simple universal forest carbon estimation method for tropical forests that reduces the amount of required field measurements. We tested this approach, along with standard regression and Random Forest modeling techniques, in a northern hardwood-dominated watershed in the White Mountains of New Hampshire. Additional objectives included assessing the effects of different inventory plot designs and GPS accuracy. The universal model performed poorly in this forested landscape due to the lack of a clear relationship between canopy height and stand basal area. Simple regression modeling also produced poor model fits; the Random Forest models produced somewhat better biomass predictions than either the universal or regression models, and had low predictive power as measured by R2 but root mean squared errors were comparable to those from other studies in complex forests. Effects of positional accuracy from survey vs. resource grade GPS units were slight, as were the effects of varying plot designs, although errors generally increased when larger basal area factors were used.