Time: 12:15 - 1:15pm Venue: City Centre Seminar Room, FB 2.07
Dr Drew Purves, Computational Ecology and Environmental Science Group, Microsoft Research Cambridge
AbstractDynamic global vegetation model (DGVM) projections differ so much that terrestrial carbon dynamics are now potentially the biggest source of uncertainty in future climate. This is because DGVMs predict carbon dynamics from representations of physiological and ecological processes that differ among models, and which are governed by functional forms and parameter values that have mostly not been estimated from, or validated against, data. So two years who, Matthew Smith and I decided to go back to the beginning, developing a simple DGVM, with an approximately equal balance of model complexity over each of the key processes governing the year-to-year dynamics of the carbon cycle. We then assembled multiple data sets on carbon stocks and flows to formally estimate, via Bayesian inference, the climate dependence of each of the processes -- and the uncertainties in these dependencies. In so doing we confirmed some dependencies (e.g. NPP), revised others (e.g. soil respiration), and identified a major qualitative uncertainty in plant mortality that strongly affects the projected range of total terrestrial vegetation carbon out to 2200. Our results demonstrate how model-data fusion can improve our understanding of Earth system processes, and lead to better constrained, more predictive, and more defensible, Earth System Models.
Chair: Dr Lisa Belyea