Spatial upscaling of process-based vegetation models: An overview of common methods and a case-study for the U.K.

Many different process-based models of vegetations are in use today. The majority of these models are parameter-rich, deterministic dynamic models, which require considerable input information and computation time. These characteristics, combined with the fact that the models tend to be parameterised at the point-support spatial scale, have made their use for larger regions problematic. Numerous examples of regional model application do exist, but how upscaling from point to region affects model output uncertainty is generally not considered. We begin by proposing a classification of upscaling methods for process-based models. Seven different methods of spatial upscaling are identified, most of which have been used in practice. We then present the example of the application of forest models to the U.K. at a 20 x 20 km grid. A discussion on upscaling uncertainty, mostly from a Bayesian perspective, concludes the paper