big questions in global change science what controls biodiversity? how will it be affected by...
TRANSCRIPT
Big questions in global change science
What controls biodiversity?
How will it be affected by climate change?
Includes students, postdocs, other faculty on campus: Pankaj Agarwal, Dave Bell, Mike Dietze, Alan Gelfand, Michelle Hersh, Ines Ibanez, Shannon LaDeau, Scott Loarie, Sean McMahon, Jessica Metcalf, Jackie Mohan, Emily Moran, Carl Salk, Rob Schick, Mike Wolosin, Hai Yu
Nature supports huge diversity
It is threatened with extinction
•Nature, 2004: 15-37% 'committed to extinction.' •IPCC: 20-30% risk extinction if temperatures rise 2°C.•Araújo: from 92% range reduction to 322% expansion.
Predicted bird losses
10%
60%
Conservation and Policy
A Framework for Debate of Assisted Migration in anEra of Climate ChangeJASON S. MCLACHLAN, *†‡ JESSICA J. HELLMANN,† AND MARK W. SCHWARTZ *
Conservation Biology 21, No. 2, 297–302
Hannah, L., Midgley, G. F., Lovejoy, T., Bond, W. J., Bush, M., Lovett, J. C., Scott, D. & Woodward, F. I.Conservation of Biodiversity in a Changing Climate. Conservation Biology 16, 264-268.
?
Guidance from science: We can’t get coexistence in models
Diversity in nature, but not in models of it
Stochasticity can help, but not much
What’s missing?
Brief history of ecological theory
1920’s to today: Systems of non-linear differential equations
- Experiments to mimic these models
1970’s to today: Forward simulation- Large models produce a mish-
mash of output- Parameterization by guesswork- Simple models with careful designs
extend analytical results
2000’s:Inferential modeling - Assimilate information- Understand more of the processes
• Insights:– Need N limiting factors to
explain N species
• Insights:– Variation can increase
diversity, but not by much– Still cannot predict diverse
ecosystems
• Hypothesis:– Many processes required to
maintain diversity– Species-specific
A role for modeling/computation
• Simple deterministic models cannot predict diverse ecosystems
• Adding stochastic elements to an otherwise simple model is not enough
• Need to better understand complexity
Challenges
• Many indirect and sparse sources of information
• Complex interactions, poorly understood
Many types of data
Experimental hurricanes
CO2 fumigation of forests
Effects of high CO2 on demography
Remote sensing
Inference on light capture by canopies
Telemetry of animal movement
QuickTime™ and aTIFF (Uncompressed) decompressor
are needed to see this picture.
QuickTime™ and aTIFF (Uncompressed) decompressor
are needed to see this picture.
Inferring pronghorn responses
Wireless sensor networks at Duke Forest
Where could a model stand in for data?
Slow variables
Predictable variables
Events
Less predictable
Molecular evidence for infection
Pathogen detection
site j
another site
another site
Host survival
environment at j
Dispersal among
sites
Transmission within sites
Host species
Pathogen taxa
An application
• Hypothesis: tradeoffs among traits needed for coexistence
• Challenge: cannot estimate the traits– They interact in unknown ways– Many types of data, all indirect
• Approach:– hierarchical Bayes inference
on all traits simultaneously
Acer trees and seeds
Experimental gaps
Demographic monitoring
Pretreatment /intervention
Spatio temporal covariates
Spatio-temporal demographic data
Individual responses with interactions
Seedbank
SeedlingImmature
treeMature
tree
maturationgermination
growth
mortality
Fecundity/dispersal
dormancy
Demography of an individual tree
Individual responses with interactions
Res
ourc
es, e
nvir
Resources, environ
A forest
Information
Seedbank
SeedlingImmature
treeMature
tree
maturationgermination growth
mortality
Fecundity/dispersal
dormancy
SeedTraps
Remote sensing:Canopy light
Covariate data:Temp, soil moisture, elevation, CO2, N
Covariates
Demographic census:Size, survival, maturation status, canopy status
Priors
Example data model Seed rain conditionally depends on all trees
500
500
1500
0 100
b) Sum with implied seed shadows
Individual seed shadows
Distance (m)
a) Liriodendron seed rain estimates
10 m contours
€
ZIP sjk,t Ajkgjk,t,β( )
€
gjk,t = fij,tK rik;u( )i=1
nj
∑
fij,t - fecundity of tree i t - yearj - plotk - seed trap samplesjk,t - seed countAjk - seed trap areagjk,t - dispersal from trees on j
fij,tK(r)
gjk,t
Latent states vs predictive intervals
€
p dX,y,D,λ( ) = p dD,λ,θ( )p θ X,y( )dθ∫
Mortality
Fecundity
Growth
Green dots are posterior means
Joint life history prediction• Parameters for process and
observation errors• Fixed year effects • Random effects (growth and
fecundity)• Latent states (canopy area,
diameter, fecundity, maturation status, mortality risk)
Evaluation
-200 yr ahead prediction has good coverage of tree-ring data- note: no age data enter model
Dashed line: 95% predictive intervalGreen lines: tree ring data
Tradeoffs among species?• Not classical tradeoffs• Within species variance large--consistent with
multiple limitations
Predictive means Individual variation
What’s ahead
Seedbank
SeedlingImmature
treeMature
tree
• A shift to inferential modeling (including prediction) – Getting the data in– Determining how things work– Finding what’s important
• Revisit analysis with new insight on how to simplify and where to retain complexity