PARENT SESSION
Organized Oral Session 47: Insights, challenges, and future directions in modeling forest dynamics at multiple scales
Organizer(s): C Tripler, C Canham, C Messier, and M Papaik
Thursday, August 11, 1:30 PM - 5:00 PM, Meeting Room 510a, Level 5, Palais des congrès de Montréal
Data assimilation, inference, and prediction in a hierarchical forest model.
Dietze, Michael*,1, Agarwal, Pankaj1, Chakraborty, Sukhendu, Clark, James1, Govindarajan, Sathish1, McBride, Allen1, Wolosin, Michael1, 1 Duke University, Durham, NC
ABSTRACT- Predicting forest response to global change requires inferential models that accommodate complex interactions, heterogeneous data sources, and scale- dependent structure. It further requires simulation tools that can efficiently address the n-body problem, the fact that model complexity scales with the square of landscape size, population size, or some combination thereof. We report on an integration of inferential and simulation tools to predict forest responses based on the full weight of evidence derived from long-term monitoring studies of mapped tree plots, remote sensing, and large- and small-scale experiments ranging from seedling plots to whole-stand CO2 fumigation and canopy manipulation. To fully accommodate uncertainty in heterogeneity of data we developed hierarchical Bayes models that allow us to infer effects of resources and climate variability on demographic rates for each of the dominant canopy species, including fecundity, dispersal, germination, growth, and mortality. Because demographic rates are estimated simultaneously and on the same individuals, we obtain parameter correlations that come from allocation tradeoffs. The full complexity is carried forward to predictive simulations that exploit graphics hardware-based computation, spatial data structures, and approximation algorithms. We demonstrate that stage structure, which provides a storage effect via seed banks and suppressed seedlings and saplings, reduces the rate of local species loss due to ecological drift compared to simple neutral models. Hubbell (2001) suggests that time to extinction decreases approximately quadratically as local community size decreases, while our simulations suggest that extinction time scales as the square root of community size. While the loss of species is very stochastic the loss of diversity (Shannon's index) shows a steady decline across community sizes from those going rapidly to monodominance to those showing no appreciable species loss.
Key words: bayesian, hierarchical, neutral model, ecological drift