Taking off the training wheels: The properties of a dynamic vegetation model without climate envelopes
Models that simulate the movement of plant types in space and time are collectively referred to as `dynamic vegetation' models. Most dynamic vegetation models use the concept of climate envelopes. These climatic limits on recruitment and survival operate in lieu of physiological understanding of the reasons why different types of plants persist in some environments where others do not. The science of quantitatively understanding plant biome boundaries is in it's infancy and so the use of climate envelopes or dominance hierarchies as a proxy for understanding is arguably a pragmatic approach to a problem of extraordinary complexity, and potentially a valid means of understanding plant distributions under altered climates. Nevertheless, interest in predicting the future of the carbon cycle has increase greatly in recent years, in concert with initiatives to collate information on plant traits and physiological functioning and increases in the sophistication of process representation in land surface models. The proliferation of alternative methods for using trait data in land surface models implicitly raises questions concerning how best to use our existing knowledge of plant processes within this new class of model. Here we present examples of trait-trade off and environmentally determined traits, to illustrate how variation in trait data, even for relatively well constrained properties, can profoundly impact model simulations.
We find that variation among leaf traits, even wihtin the observed range, and minor modifications to assumptions governing carbon cycling processes, can have dramatic, continent-scale impacts on the modelled position of biome boundaries in our trait-based model. We advocate for the use of both parametric and structural ensembles when designing and analyzing models with this degree of complexity, and for the continued improvement and analaysis of of trait databases within the context of vegetation dynamics models.