Comparing models for tree distributions: concept, structures, and behavior

Abstract

Vegetation models are increasingly used for assessing impacts of a changing environment on landscapes. Model evaluation is an important task since it allows us to determine of model accuracy and applicability. Rarely, model evaluation includes comparisons with model outputs from approaches that are based on completely different ecosystem-theoretical concepts. However, such comparisons are important for improving the understanding of model behavior. In this paper, two model concepts the static equilibrium and the dynamic-transient are compared using simulations of tree distributions. The simulations were compared against empirical data of the Swiss National Forest Inventory (NFI). Static equilibrium models (e.g., regression model) simulate tree distributions primarily as a function of the abiotic biophysical environment. The approach can be considered as ‘top-down’ since the data used to calculate the abiotic environment integrate over large spatial scales up to km 2, and hence mirror major features of the tree habitat, but does not describe individual physiological properties of the tree species. Such models are calibrated with empirical data sets, and thus the resulting simulations can be considered to mirror realized niches that account for management. The dynamic-transient approach (e.g. ForClim gap model) simulates trees in a ‘bottom-up’ approach since detailed species-specific local-scale life-history attributes and environmental variables are considered to describe the trees. As management schemes are not intrinsic to ForClim, resulting simulations may be viewed as realized niches in the absence of management. Results of the model comparison show that at large spatial scales both models discriminate well between major tree distribution characteristics and can be considered as valid estimators for assessing regional vegetation patterns. Specifically, the model comparison generated valuable insights into human-induced alterations of species–specific distribution patterns. For example, simulations of the regression model and observations of the NFI agreed to a large extent regarding the distribution of Fagus sylvatica. ForClim, however, clearly overestimated this species in most ecoregions. Thus, these results suggest that F. sylvatica would be more frequent if management effects had not reduced its range; similar conclusions can be drawn in other respects, as discussed in the paper.