Comparison of Forest Inventory and Analysis surveys, basal area models, and fitting methods for the aspen forest type in Minnesota

The Forest Inventory and Analysis (FIA) unit of the U.S. Forest Service has collected, compiled, and made available plot data from three measurement periods (identified as 1977, 1990, and 2003, respectively) within Minnesota. Yet little if any research has compared the relative utility of these datasets for developing empirical yield models. This paper compares these and other subdatasets in the context of fitting a basal area (B) yield model to plot data from the aspen (Populus tremuloides Michx.) forest type. In addition, several models and fitting methods are compared for their applicability and stability over time. Results suggest that the three parent datasets, along with their subdatasets, provide very similar three parameter B yield model prediction capability, but as model complexity increases, variability in coefficient estimates increases between datasets. The absence of data for older aspen stands and the inherent noise within B data prevented the exact determination of an overall best model. However, the model B = b₁S^b₂(1 − exp( − b₃A)) with site index (S) and stand age (A) as predictors was found consistently among the highest in precision and stability. Additionally, nonlinear least squares and nonlinear mixed-effects fitting procedures produced similar model fits, but the latter is preferred for its potential to improve model projections. The results indicate little practical difference between datasets from different time periods and different sizes when used for fitting the models. Additionally, these results will likely extend to other states or regions with similar remeasurement data on aspen and other forest types, thus facilitating the development of other ecological models focused on forest management.