Subject: A training workshop for Macrosystems research Integrating evidence on forest response to climate change: physiology to regional abundance. A training workshop for Macrosystems research 13-14 May 2013, Thomas Center, Duke University
At the 2012 annual workshop for the NSF Macrosystems research program we discussed training opportunities for advanced graduate students and postdoctoral studies in eco-informatics and modeling. Our own macrosystems project has several leader in this area who have agreed to present a training workshop. We can invite a small group of grads/postdocs for this activity that will take place at Duke University on 13-14 May 2013. Methods focus on synthesis of evidence for the effects of climate change on forest processes. Goals of the workshop are to engage state-of-the-art experimental and observational data analyses from physiological to species distribution modeling in the development of synthetic models. We can train a small group (8 to 12) of advanced graduate students and post-docs. Presentations for the 2-day training activity will combine modeling and computation, including hands-on experience with software in R. Participants will be asked to cover travel expenses. We can cover costs of the workshop, including lodging and meals. To apply please send to [email protected] a current cv and a paragraph describing 1) your background in quantitative methods and 2) your thoughts on how this workshop could benefit your own research. We will accept applications through 15 Apr, but will give preference in order of receipt. Agenda Sunday, 12 May pm arrival and reception Monday, 13 May 8:30 Clark Welcome, logistics, overview; summarize motivation, goals of the 2-day workshop, schedule of activities am session Clark Individual scale, regional consequences: the macrosystems approach State space modeling of demographic rates from tree census data pm session Finley/McMahon Spatial modeling of forest attributes using high-dimensional remotely sensed data and field inventories group projects pm group dinner Tuesday, 14 May am session Dietze Integrating forest data into ecosystem models pm session Gelfand Spatial scaling of integral projection models: individuals to populations group session and wrap up Methods workshop abstracts: Clark The two goals of this session are 1) to introduce the motivation for individual scale analysis for regional scale interpretation, and 2) to discuss modeling and computation issues for hierarchical dynamic models of demography. The modeling topics include structure and assumptions, use of prediction to evaluate models and identify important input variables. Computation will be included using software in R. Dietze This session will focus on how to apply Bayesian approaches to the calibration of processes-based ecosystem models and the propagation of uncertainty into model forecasts. Building on the previous sessions, we will focus on the integration of forest inventory, tree ring, and remote sensing data. In addition, we will utilize the PEcAn workflow system and R to apply these methods to a simple ecosystem model (SIPNET) and a sophisticated terrestrial biosphere model (ED2). Finley/McMahon The session will blend modeling, computing, and data analysis using a variety of LiDAR and experimental forest inventory data. We will briefly cover LiDAR and forest inventory data preprocessing using R, hierarchical spatial Bayesian model specification, parameter estimation, prediction, and inference. Some advanced topics will include multivariate models with missing data and settings where the number of observations is too large to efficiently fit the desired hierarchical models. Special attention will be given to exploration and visualization of data and the practical and accessible implementation of spatial models using R and lower-level programming languages. Gelfand Integral projection models (IPMs) are used to estimate demographic functions (survival, growth, fecundity). The approach does not align projected distributions with the observed data, and it introduces an inherent mismatch in scales. Parameter estimates on individuals do not allow population-level interpretation. I discuss a new three-stage hierarchical model to infer dynamics within a Bayesian framework. Exact Bayesian model fitting is computationally challenging; we offer approximate strategies to facilitate computation. We illustrate with simulated data examples as well as well as a set of annual tree growth data from Duke Forest in North Carolina.
