Integrating large-scale survey data sets with climate and land use data to model species distribution dynamics

1
Integrating large-scale survey data sets with
climate and land use data to model species
distribution dynamics
Biodiversity Over Space

• Andrew M. Latimer and John A. Silander
• Department of Ecology Evolutionary Biology
• University of Connecticut

2
distribution and abundance of species
3

• Data-model integration
• New large data sets
• Issues
• Diverse kinds and scales of data
• Spatial and temporal covariance structures
• Tools Bayesian hierarchical models

4
In this talk

• Static single-species distribution models
• Areal-unit models
• Environment and colonization
• Land use
• Abundance
• Biodiversity joint distribution of species
• Point process models
• Computational advances
• Multispecies models
• Current work making these dynamic

5
Species observations
Abiotic environment
Land use data
Presence/absence Abundance Diversity

• Why absent?
• Not suitable
• Not available
• Not colonized

E(Abundance) P(Presence)?
6
Hierarchical single-species model
Environmental data (weather stations, soils)
P(suitable)
Land use data (satellite imagery)
P(available suitable)
Neighborhood connectivity info
P(colonized suitable available)
Species sample data
P(present)
7
Hierarchical single-species model
P() dbin(ni, pi) logit(pi) XTß wi
P(suitable)
P() f(Ui) 1-Ui where Ui prop.
human-altered
P(available suitable)
P() dbin(ni, qi) logit(qi) g(neighborhood)
P(colonized suitable available)
Latimer et al. (2006) Ecological Applications
8
Likelihood
P(present)

• Where the species was observed
• L(yi) Binomial(ni, qi) f(pi)
• Where not observed
• L(yi) (1-f(pi)) f(pi)(1-qi)ni

Probability present given suitable available
Suitability, adjusted by availability function
Probability unsuitable and/or unavailable
Probability suitable available but not observed
9
White Proteas (Protea spp.)
P. lacticolor P. aurea P. punctata
P. punctata
P. aurea
P. lacticolor
10
Hierarchical model P(suitable)
11
Hierarchical model P(available)P(suitable)
12
Hierarchical P(colonized suitable available)
13
Inference

• Primarily environmental limit on presence
• Some constraint at colonization stage

14
Adding an abundance level
P(present)
P(abundance(k) present)
Ordinal abundance scores, environmental data
Introduce latent (log-scale) abundance surface
Z and cutpoints c1, c2, , ck. Abundance score
1 if zi c1 2 if zi gt c1 and zi
c2 k if zi gt ck
15
Latent log-scale abundance surface (Z)
16
Inference

• Different factors drive abundance
• Cool winter temperature vs warm wet growth
  season
• Mechanism?
• Germination vs growth

Latimer et al. Oecologia (in review)
17
Multi-species models
Potential richness in the absence of human
landscape alterations
Adjusted (transformed) richness given human
transformed landscapes
Gelfand et al. (2005) Bayesian Analysis
18
Modeled subregion (for a subset of 40 species)
Computational issues Please help!
19
Species Richness
20
Multi-species results

• Different land uses differential impacts.

Latimer et al. (2004) S.A. Journal of Science
21
Modeling with point data
Curse of dimensionality
Predictive process approach Banerjee, Gelfand et
al.
22
Multiple spatial processesMulti-species model
Cel. orbiculatus
Berberis thungbergii
Rosa multiflora
Euonymus alatus
23
Canopy Closure
Berberis thunbergii Celastrus orbiculatus Rosa
multiflora Euonymus alatus
Density
Regression coefficient value
24
Celastrus orbiculatus
Berberis thungbergii
Sample data
present
absent
Rosa multiflora
Euonymus alatus
25
Summing up

• Opportunities
• Physiological responses, abundance structure
• Land use change impacts
• Integrating satellite data
• Limitations
• Colonization and other spatial factors
• Computer power

26
Current work

• Making dynamic climate change
• Data population-level performance data over time
• Field survey plants and populations over
  gradients
• Satellite data for phenology and productivity

Latimer Wilson et al. (in prep.) Global Change
Biol
27
Acknowledgments

• U.S. NSF Grant DEB 0516320
• SANBI (Esp. Tony Rebelo Guy Midgley)
• Duke ISDS (Esp. Alan Gelfand Huiyan Sang)
• UCONN EEB (Esp. Inés Ibáñes, Adam Wilson)