On the Ecology of Mountainous Forests in a Changing Climate: A ...

Introduction 9
(1984) provides a more detailed description of the common characteristics of forest gap
models.
During the last 20 years, many forest gap models have been developed based on the
parent model JABOWA, which was built to simulate succession in a northern hardwood
forest of the eastern United States (Botkin et al. 1970, 1972a,b; the name of the model
stands for the three authors, F. JAnak, D. BOtkin & J. WAllis). The aim of their study
was “to introduce a minimal number of assumptions and to find the simplest mathematical
expression for each factor that was consistent with observations.” (Botkin et al. 1972a,
p. 850). They were remarkably successful in that respect, but the model was fairly
expensive to run given the computer resources of that time.
The adaptation of JABOWA for southern Appalachian forests led to the model FORET
(Shugart & West 1977), which was equally successful in predicting the effect of a fungal
disease (the chestnut blight) on forest composition. Subsequently an amazing proliferation
of forest gap models took place: Models were developed for tropical forests
(Doyle 1981), forests in Australia (Shugart & Noble 1981), in the western United States
(Kercher & Axelrod 1984), in Central Europe (Kienast 1987), and in the boreal zone
(Leemans & Prentice 1989, Bonan & van Cleve 1992, Shugart et al. 1992). Moreover,
the approach seems not to be restricted to forests: Smith et al. (1989) and Coffin &
Lauenroth (1990) successfully developed gap models for grasslands. Thus, the gap dynamics
hypothesis proved to be a viable concept in a wide variety of ecosystems. It is
also remarkable that these models are closely related to each other: Many of the equations
formulated for JABOWA more than 20 years ago are still being used today without modification
(Botkin 1993).
Parallel to the adaptation of forest gap models for various ecosystems, ever more details
were added to these models, such as nitrogen availability and nutrient cycling (Aber et al.
1979, 1982, Aber & Melillo 1982, Weinstein et al. 1982, Pastor & Post 1985), the influence
of fire (Kercher & Axelrod 1984), ecological indicator concepts (Kienast 1987),
seed dispersal by birds (Keane et al. 1990), herbaceous vegetation (Kellomäki &
Väisänen 1991), and detailed biophysical-ecophysiological submodels (Martin 1990,
1992, Bonan & van Cleve 1992, Friend et al. 1993). However, the increasing complexity
of forest gap models made simulation studies ever more tedious and precluded detailed
model analyses. For example, current models typically include 1000 to 1500 parameters
(Shugart 1984, Kienast 1987); hence, an all inclusive sensitivity analysis is almost prohibitive.
Not surprisingly, only few sensitivity studies have been conducted, covering