Using sensitivity and uncertainty analyses to improve predictions of broad-scale forest development

Abstract

The FORET model of forest development can be used to estimate influences of natural and anthropogenic disturbances on forest dynamics. Interpreting the effects of these disturbances requires an understanding of sources of error in the model; however, error analysis of a complex model is not straightforward. The method of applying sensitivity and uncertainty analyses in this paper takes advantage of the modular structure of the FORET code to examine errors that occur as a result of parameter variability or inadequate representation of the natural variability of the system. The FORET model can be broken down into factors affecting tree birth, death and growth, and the growth module is composed of an equation for optimal growth that is modified by effects of prevailing environmental conditions. Both the analytical and Monte Carlo sensitivity analyses show that optimal tree growth is most sensitive to the maximum diameter of a species (a parameter obtained from field measures). The uncertainty analysis focuses on spatial heterogeneity as a source of errors due to inadequate representation of the natural variability. Errors that occur as a result of extrapolating from a few plots to a region have greater impact on projections of above-ground stem-wood volume for those forest types which have the highest projected volume per area and the greatest land area covered. Measured and projected forest volume for the state of Vermont compare well. Improvements in the comparison depend on reducing the variability in measured volume, particularly for the maple-beech-birch forest type. These results demonstrate procedures of sensitivity and uncertainty analyses of complex simulation models that are useful for interpretation of model projections for large regions.