Forest Models Defined by Field Measurements: Estimation, Error Analysis and Dynamics

Abstract

A spatial and mechanistic model is developed for the dynamics of transition oak—northern hardwoods forests in northeastern North America. The purpose of the model is to extrapolate from measurable fine—scale and short—term interactions among individual trees to large—scale and long—term dynamics of forest communities. Field methods, statistical estimators, and model structure were designed simultaneously to ensure that parameters could be estimated from data collected in the field. This paper documents eight aspects of a three—year study to calibrate, test, and analyze the model for the nine dominant and subdominant tree species in transition oak—northern hardwoods forests: 1) Design and structure of the model. The model makes population dynamic forecasts by predicting the fate of every individual tree throughout its life. Species—specific functions predict each tree's dispersal, establishment, growth, mortality, and fecundity. Trees occupy unique spatial positions, and individual performance is affected by the local availability of resources. Competition is mechanistic; resources available to each tree are reduced by neighbors. Although the model was developed to include light, water, and nitrogen, the version described here includes only competition for light (shading and light—dependent performance) because the field data provide little evidence of competition for nitrogen and water over the range of sites examined. 2) Estimates of the model's parameters for each species. The estimates reveal a variety of "strategic trade—offs" among the species. For example, species that grow quickly under high light tend to cast relatively little shade, have low survivorship under low light, and have high dispersal. In contrast, species that grow slowly under high light tend to cast relatively dark shade, and to have high survivorship under low light and low dispersal. These trade—offs define one of two dominant "axes" of strategic variation. 3) Community level predictions of the model. The model predicts succession from early dominance by species such as Quercus rubra and Prunus serotina, to late dominance by Fagus grandifolia and Tsuga canadensis, with Betula alleganiensis present as a gap phase species in old—growth stands. The model also predicts that old—growth communities will have intraspecifically clumped and interspecifically segregated spatial distributions. 4) An error analysis that identifies community level predictions that are robust given the level of sampling uncertainty in the study. This analysis translates the statistical uncertainty associated with each parameter estimate into statistical uncertainty in the model's predictions. The robust predictions include those mentioned in aspect (3) above. 5) Sensitivity of the model to changes in initial conditions and to changes in the three parameters not included in the error analysis. For example, the model predicts that initial abundances continue to affect community composition well into succession (> 300 yr for some species). 6) Tests of the system— and community—level predictions of the model against independent data gleaned from other studies. These tests support the predictions found to be robust in the error analysis, including those predictions mentioned in aspect (3) above. 7) Modeling experiments that determine which aspects of individual performance and inter—neighbor competition are responsible for each of the robust predictions identified in aspect (4) above and tested in aspect (6) above. This analysis reveals a wide variety of causal relationships, with most parameters contributing to at least one community level phenomenon. 8) An explanation of the diversity of individual level causes identified in aspect (7). The two "axes" describing most of the strategic variation among the species (see [2]), provide a simple explanation of community level pattern in terms of individual level processes.