Discretizing diversity: What do two decades of plant functional ecology tell us about parameterizing plants in vegetation models?

Friday, December 13, 2019 - 13:40
Add to Calendar

Earth System Models typically represent the ~1/2 million terrestrial plant species with 8-20 ‘Plant Functional Types’ (PFTs) based loosely on plant structure and biome type. Next-generation biosphere models require more sophisticated representations of plant diversity to capture complex spatial and temporal ecosystem dynamics. Yet vegetation parameterization represents a highly underconstrained problem in which equifinality (many parameter combinations yielding similar model behavior) is a constant challenge, and constraining model parameters with actual observations of plant functional traits remains difficult. Emerging ecological work suggests that many important plant traits are phylogenetically conserved (closely related species tend to have related traits). Thus, evolutionary lineages may represent a unified, theoretically defensible and scalable framework for representing plant diversity through ‘Lineage Functional Types’ (LFTs). By parameterizing functional types based on evolutionary lineage rather than structural similarity, modelers could consistently modify the complexity of plant type discretization for specific applications using phylogenetic relationships. At the same time, phylogenetic relationships could allow existing databases of model-relevant plant traits to be combined with plant inventories to constrain model parameters for different LFTs. We illustrate the power of the LFT concept using data from forests in the Pacific Northwest, USA and highlight how recent ecological insights can be brought to bear on the problem of model parameterization. Many Earth System Models are moving to more complex representations of vegetation structure that include competition for light, nutrients, and water. These competitive interactions depend on plant traits; robustly disaggregating vegetation to represent these functional differences could enhance model predictions.

Link for More Information: