Transcript Terrestrial modeling group - Statistical and Applied

Terrestrial modeling group
Benoit Courbaud (visiting faculty)
Jim Crooks (Samsi postdoc)
Mike Dietze (Harvard)
Cari Kauffman (Samsi postdoc)
Sean McMahon (Duke postdoc)
Jonty Rougier (Samsi visitor)
Wei Wu (Duke postdoc)
Jim Clark (Duke)
Challenges
• Infer forest diversity response to climate
change
• Demographic rates of all species depend on
interactions with other species, resources,
and climate
• Inference:
– Hidden variables
– Data at many scales
• Prediction: large number of interactions
– Regional climate affects diversity at tree scales
– Climate drivers: GCMs need down-scaling
Beginnings
• Stand simulator parameterized to longterm data sets
• A SAMSI “terrestrial modeling group”
• Goals:
– Downscale climate to local soil moisture
(Cauffman, Rugier, Wu)
– Model emulation (Crooks)
Hierarchical Bayes model
Data
Seed
traps
CO2
treatment
Maturity
obs
Climate
Diameter
data
Canopy
data
TDR
Survival
Soil
moisture
Light
Dispersal
Maturation
Diameter
Fecundity
Diameter
growth
Mortality
risk
Processes
Observation errors
Process uncertainty
Parameters
Hyperparameters
Heterogeneity
Climate through soil moisture
Many types of data and
models
Inference at the individual tree level
28
10
11
12
29
30
31
44
13
45
46
47
‘mean response’
0
20
40
Diameter (cm)
Clark, LaDeau, Ibanez
Ecol Monogr (2004)
60
80
100
year effect
random individual
effect
error
120
Demographic rates depend on soil
moisture
Cari to discuss soil
moisture modeling
Stochastic Emulation
●
●
The forest simulator has properties that make emulation
both important and difficult:
–
Slow speed limits the physical area that can be simulated in
reasonable time.
–
Its output is
–
stochastic
–
non-gaussian
–
varies over the input space.
Need for a nonparametric statistical method to emulate the
entire output distribution across in the input space.
Stochastic Emulation
• Use the Kernel Stick-Breaking
Processes idea of Dunson and Park
(2006) to build such an emulator.
• No final results yet, but the methods and
preliminary work will be presented
tonight in Jim Crooks' poster.
Our ‘transition workshop’
• April 2007
• Climate scientists and ecologists discussed
applications to a range of temperate and tropical sites
• Discussion of Research Coordination Network
proposal
• Using data from a range of sites, can we more fully
explore implications of climate change for forest
biodiversity?