Incorrect representation of uncertainty in the modeling of growth leads to biased estimates of future biomass.

Biomass is a fundamental measure in the natural sciences, and numerous models have been developed to forecast timber and fishery yields, forest carbon content, and other environmental services that depend on biomass estimates. We derive general results that reveal how dynamic models that simulate growth as an increase in a linear measure of size (e.g., diameter, length, height) result in biased estimates of future mean biomass when uncertainty in growth is misrepresented. Our case study shows how models of tree growth that predict the same mean diameter increment, but with alternative representations of growth uncertainty, result in almost a threefold difference in the projections of future mean tree biomass after a 20-yr simulation. These results have important implications concerning our ability to accurately predict future biomass and all the related environmental services (e.g., forest carbon content, timber and fishery yields). If the objective is to predict future biomass, we strongly recommend that: (1) ecological modelers should choose a growth model based on a variable more linearly related to biomass (e.g., tree basal area instead of tree diameter for forest models); (2) if field measurements preclude the use of variables other than the linear measure of size, both the mean and other statistical moments (e.g., covariances) should be carefully modeled; (3) careful assessment be done on models that aggregate similar individuals (i.e., cohort models) to see if neglecting autocorrelated growth from individuals leads to biased estimates of future mean biomass.