Predictive modeling of vegetation distributions

1
Predictive modeling of vegetation distributions

• Symposium on Bioinformatics Temporal and Spatial
  Syntheses of Vegetation Data
• International Association of Vegetation Science
• 49th Annual Meeting, Palmerston North, New
  Zealand
• 12-16 Feb 2007
• Janet Franklin
• Vegetation Science Landscape Ecology Laboratory
• Department of Biology
• San Diego State University

2
Acknowledgements

• US National Science Foundation (0452389)
  Geography Regional Science Program
• Jennifer Miller, West Virginia University
• Robert Taylor, US National Park Service, VTM data
  champion
• Tom Edwards, Mike Austin, Kim van Neil and many
  others

3
Outline

• Introduction
• What is Species Distribution Modeling (SDM)?
• What is special about vegetation data?
• Framework for SDM
• The Data Model and Vegetation Data
• Sample design
• Response variable
• Explanatory environmental variables
• Scale

4
What are species distribution models?

• Quantitative models of species-environment
  relationships
• used to predict the occurrence of a species for
  locations where survey data are lacking
  (interpolate biological data in space)
• Species abundance or presence
• Habitat suitability
• Realized niche

5
What do you need?

• data on species occurrence in geographical space
• maps of environmental variables
• A model linking habitat requirements to
  environmental variables
• A way to produce a map of predicted species
  occurrence -- GIS
• Data to validate the predictions

6
The Data
Elevation, Quercus pacifica Presence (n131),
Absence (n797)
7
Potential Solar Radiation (winter solstice)
8
Probability of Species Presence
Channelislandsrestoration.com
9
Why make spatial predictions of species
distributions?

• Conservation planning
• Reserve design
• Impact assessment
• Land and resource management
• Climate change
• Invasive species
• Ecological restoration
• Population viability analysis
• Modeling community dynamics

10
What is Special About Vegetation Databases and
Databanks?

• Lots of it
• Multiple species (community)
• Presence and absence, abundance
• Plants not (usually) (very) cryptic or mobile
• - May come from multiple surveys
• - Time periods may vary
• - Protocols may vary
• - May lack locational precision

11
Wieslander California Vegetation Type Mapping
Survey -1930s
18,000 plots state-wide 1481 Southern California
shrubland plots 400-m2, 233 species
(http//vtm.berkeley.edu/)
Los Angeles
San Diego
12
Framework for Modeling Species Distributions
Ecological Model
Data Model
Empirical Model

• Any mechanistic process model of ecosystem
  dynamics should be consistent with a static,
  quantitative and rigorous description of the same
  ecosystem (Austin 2002, p. 112)

13
The Data Model

• Theory and decisions about how the data are
  sampled and measured
• Sampling in space and time
• Response variable
• Predictor variables
• Spatial scale
• Resolution
• Extent

14
Sampling in Vegetation Surveys

• - Not always probability-based
• But
• dense data can be sampled
• can supplement with random sample

Yucca brevifolia Alliance Pr/Abs
15
Response Variable in Vegetation Surveys

• Presence or abundance of all plant species makes
  it possible to
• Model species
• Model communities
• Predict (species) first, then classify
• Classify or ordinate (community) first, then
  predict
• (review of modeling communities by Ferrier and
  Guisan 2006 J. Appl Ecol 43393-404)

16
SDM is direct gradient analysis
Fundamental vs. realized niche
Resource utilization function
Date from John T. Curtis. Figure from Gurevitch
et al. The Ecology of Plants
17
Model species first, then classify community

• Vegetation continuum, composition varies
  continuously, individual species responses to
  gradients (Austin 1998 AMOB 852)

Ferrier et al. 2002, Biodiv. Conserv 112309
18
Classify first, then model

• Predictive Vegetation Modelling (Franklin 1995
  Progr Phy Geogr)

Yucca brevifolia Alliance Pr/Abs
19
Ordinate and model together (CCA)

• Oregon coastal ranges, forest (800 plots,
  multiple surveys and agencies)
• (Ohmann and Gregory 2002 Can J For Res)

20
Classify or ordinate first, then model(or
classify and model together)

• Classify first, then model starts with indirect
  gradient analysis of communities
• Classify/ordinate and model environment together
  is direct gradient analysis of communities

21
Summary Vegetation Surveys and Databanks

• Are large datasets, often geographically
  comprehensive
• Can overcome some sampling problems
• New modeling methods robust to data quality

22
Summary Vegetation Surveys and Databanks

• Usually include P/A or abundance of all plant
  species
• P/A data yield powerful species models
• ? Community composition data may be underutilized
  in vegetation modelling

23
Thank you!Questions?
24
What do we really want?
25
Plant Distributions Primary Environmental Regimes
Guisan Zimmerman (2000)
26
Predictor Variables for Vegetation Modelling
Slope Curvature
Solar Radiation
27
Scale in Species Distribution Modeling

• Biogeographical scale
• Point observations
• Lots of them
• Not from designed surveys
• Presence only, atlases, collections
• Resolution of analysis 10x10-50x50 km
• Many to one
• Ecological scale
• Scale of data collection 102-103 m2
• Probability sample designs
• Resolution of analysis 10x10 to 1000x1000 m
• One to one

McPherson et al. (2006)
28
Biogeographical Scale
Assessment of Potential Future Vegetation Changes
in the Southwestern United StatesRobert S.
Thompson, Katherine H. Anderson,, Patrick J.
Bartlein
http//geochange.er.usgs.gov/sw/impacts/biology/ve
g_chg_model/
29
Scale in Species Distribution Modeling

• Biogeographical scale
• Point observations
• Lots of them
• Not from designed surveys
• Presence only, atlases, collections
• Resolution of analysis 10x10-50x50 km
• Many to one
• Ecological scale
• Scale of data collection 102-103 m2
• Probability sample designs
• Resolution of analysis 10x10 to 1000x1000 m
• One to one

30
Ecological Scale
Channelislandsrestoration.com
31
Biogeographical scale
Ecological scale
32
Summary Vegetation Surveys and Databanks

• Plant distributions primarily controlled by
  light, heat sum, water and nutrients
• Tools and data exist for mapping environmental
  gradients related to these primary regimes

33
Summary Vegetation Surveys and Databanks

• Modeling and spatial prediction at
  biogeographical or ecological spatial scale
• Coarse-scale modeling can overcome locational
  errors in historical surveys
• - But limited to coarse-scale predictors
  (climate, not terrain)

34
Conceptual model of geographical data(Goodchild
1994)

• Field geographical space is a multivariate
  vector field where variables can be defined and
  measured at any location
• Elevation
• Vegetation type
• Entity empty geographical space contains objects
• Tree
• Species occurrence
• Fire perimeter

35
The Species Data Model

• In species distribution modeling we start with
  entities
• observations of species occurrence
• and end with fields
• Maps of probability of occurrence

36
What do we really want?
San Diego County is 11,721 km2 San Diego Bird
Atlas http//www.sdnhm.org/research/birdatlas/yel
lowwarbler.html