Caught in the flux net: disentangling error, uncertainty, heterogeneity, and spatial process in biogeochemical scaling

Abstract:

Attempts to link observations across multiple scales, and in particular the problem of scaling up fine-scale observations to landscape and regional process, faces numerous theoretical, computational, and statistical challenges. This talk aims to link theoretical advances in the scaling of biotic heterogeneity, abiotic heterogeneity, and contagious disturbance with statistical advances for linking observations that integrate over different scales. Critical to this goal is the need to partition sources of uncertainty and variability. In particular, the variance you can calculate most readily is rarely the most important or relevant one to quantify. In community ecology, hierarchical Bayes (HB) latent-variable models that separate true variability in ecosystem processes from observation errors have challenged long-standing theory surrounding the maintenance of biodiversity, yet application of such approaches to regional-scale biogeochemical processes is just beginning. A special case of such models, focused on the change of support problem, deal specifically with linking observations that integrate over different spatial and temporal scales.

Occurring in parallel with these statistical advances have been the development of new theories for the spatially-implicit scaling of biotic and abiotic heterogeneity, as well as contagious disturbances such as fire and pathogens, and the incorporation of such approaches into process-based ecosystem models. Such approaches upscale by integrating the probability distributions of system heterogeneity over the functional response of the ecosystem to such heterogeneity. We demonstrate that this approach can also be downscaled by conditioning on incomplete partial observation at a local scale, and can have lower uncertainty than brute-force spatially-explicit approaches. We also extend such approaches from integrating over observed heterogeneities to integrating the latent, high-dimensional variability in HB models. Finally, there is a pressing need to quantify the important sources and scales of variability and uncertainty in ecosystem processes that affect our ability to forecast ecosystem processes.