Vegetatio 86: 1-13, 1990.
© 1990 Kluwer Academic Publishers. Printed in Belgium l
Evaluating performance of an Appalachian oak forest dynamics model
Elizabeth A. Harrison I & Herman H. Shugart
Department of Environmental Sciences, University of Virginia, Clark Hall, Charlottesville, VA, 22903,
USA; ~Current address: Department of Geology, Michigan Technological University, Houghton,
MI 49931, USA
Accepted 14.2.1989
Keywords: Biological model, Biomass, Model testing, Parameterization, Simulation model
Abstract
An Appalachian-oak forest in a small montane watershed was sampled to calibrate and test an existing
forest dynamics simulation model. Indices developed in earlier studies by Whittaker were used to estimate
the response of different tree species to soil moisture. As is the case in many forest modeling applications,
neither detailed environmental data at a micro-spatial scale nor quantitative historical stand data were
available for the study.
A protocol of both model parameter estimation and simulation output evaluation is developed. The
method involves simulation experiments under which the model parameters are allowed to vary systemati-
cally with respect to environmental control variables, allowing one to generate a field of potential
simulation results that can be searched for patterns shared by the observed data. The degree to which
common patterns emerge provides a measure of model sensitivity to patterns at the scale of interest. This
protocol can provide an appraisal of the appropriateness of a simulation model to the system of interest
and can be used to assess the performance of the model in future applications.
Nomenclature: Little, E.B. 1971. Atlas of United States trees. Government Printing Office, Washington,
D.C.
Introduction
Gap models (Shugart & West 1980; Shugart
1984; Dale etaL 1987) are a subclass of indi-
vidual-tree-based models (Munro 1974) of forest
dynamics that have been used to investigate suc-
cession in natural forests. The development and
testing of these models has often focused on
regional-scale forest pattern (Solomon 1986;
Shugart et al. 1980), and long-term (hundreds of
yr) successional dynamics (Shugart 1984). In
typical model tests, the average behavior of 50 to
100 of the small (ca 0.1 ha) plots simulated by the
models is inspected for decade- to century-
duration forest responses. In many cases the time
window of application is quite long (e.g., predic-
tion of old-growth forest composition, Leemans
& Prentice 1987; model reconstruction of fossil
pollen chronologies, Solomon 1986).
There is an increasing interest in model applica-
tions that involve site-specific stand projections
over relatively short periods of time. These appli-
cations imply use of the models at spatial and
temporal scales for which they are largely