Download Ecological Niche Modeling: Concepts and Applications

Ecological Niche Modeling:
Concepts and Applications
A. Townsend Peterson
University of Kansas Natural History Museum
Niche Concepts
Many definitions exist for ‘niche’ …
I avoid those that are process-oriented (e.g., “the
role of species X in its community”)
Focus on a more workable definition: “the set of
environmental conditions within which a species
can maintain populations without immigrational
subsidy” (J. Grinnell)
Grinnellian concept has the advantage of a
single focus (environmental conditions)
Interactive (Eltonian) processes probably act
chiefly on finer spatial scales and are not
commonly discernable on coarse scales
The BAM Diagram
Abiotic niche
Area
presenting
appropriate
combinations
of abiotic and
biotic
conditions (=
potential
distribution)
Actual geographic distribution
(abiotic and biotic conditions fulfilled,
accessible to dispersers)
Accessibility
Biotic interactions
Species Interactions Important
Abiotic niche
Accessibility
Biotic interactions
Species Interactions Unimportant
Abiotic niche
Biotic interactions
Accessibility
Niche Modeling
Physiological constraints Ecological
processes Geographic phenomenon
As such, geographic phenomena of
distributions should be reconstructed in
ecological spaces
Linked spaces, in which there is a one-toone mapping between elements in G and
elements in E
Modeling best carried out in E
Easy-to-measure
variables
(scenopoetic)
Proximate
variables
(scenopoetic
and/or
bionomic)
S1
S2
P1
S3
P2
S4
P3
S5
P4
S6
Presence or absence
of suitable conditions
for the species in
question
“Presence”
Absence Data
Abiotic niche
Accessibility
Biotic interactions
Niche Modeling Considerations
Input Data
Presence data should be
unbiased in sampling
environments
OR interpretation limited to
environments sampled
Absence data are of
dubious utility
Environmental data must be
proximate … i.e., as close
to causal as can be
Occurrence data and
environmental data must be
of comparable resolution
spatially and temporally
Modeling Considerations
Enough complexity to
capture details of niches
Not too much complexity to
avoid overfitting
Model response types must
be appropriate to the
complexity of ecological
niches
Appropriate regions for
training and testing models
must be chosen based on
M (accessibility)
Model validation is not
simple
Major Findings
ENM offers a means of identifying nonrandom
associations between species’ occurrences
and ecological features (= niche)
Niches are conserved over ecologicalevolutionary time periods
Niches are conserved even in novel
community contexts
ENM offers excellent predictivity of geographic
phenomena related to biodiversity …
Niche Conservatism I:
Invasions are Predictable
Aedes albopictus
Aedes albopictus
Known as the “Asian
Tiger Mosquito”
Invader; fastest
spreading mosquito in
the world
Aggressive daytime
biter and pest
Known to transmit
Dengue, La Crosse, St.
Louis, Eastern Equine,
Ross River, Rift Valley,
and West Nile Viruses
Abiotic niche
Area
presenting
appropriate
combinations
of abiotic and
biotic
conditions (=
potential
distribution)
Actual geographic distribution
(abiotic and biotic conditions fulfilled,
accessible to dispersers)
Accessibility
Biotic interactions
Abiotic niche
Area
presenting
appropriate
combinations
of abiotic and
biotic
conditions (=
potential
distribution)
Actual geographic distribution
(abiotic and biotic conditions fulfilled,
accessible to dispersers)
Accessibility
Accessibility
Biotic interactions
Aedes albopictus
Present predicted distribution, native range in Asia
Aedes albopictus:
USA invasion
Projected Asian niche into USA present to create
invasion risk-map. How well did GARP perform...
Aedes albopictus:
USA invasion
Aedes albopictus:
world risk-map
Conservatism II
Predictivity Across the End of the
Pleistocene
Pollen-based Analyses
Picea sp.
Brasenia schreberi
Acer saccharum
Niche Conservatism Evidence
Short term – invasions are predictable in
terms of potential geography
Middle term – longitudinal studies
Pleistocene to Present are successful
Middle-to-Long term – sister species tend
to be very similar
Long term – tracing over phylogeny often
(but not always) shows conservative
evolutionary patterns
Applications II: Marine Intrusion and
Climate Change Effects on Biodiversity
Coastline Topography and Marine
Intrusion
New Zealand - Coromandel Coast
South Coast New Guinea
The Americas
Global Projected Extinctions from
Marine Intrusion
Global Species Losses:
181 species under the 1 m scenario
337 species under the 6 m scenario
out of 18,628 current species considered
Mexican Sheartail Doricha eliza
Joint Effects
Double whammy
More species affected
Joint Effects
Percent loss owing to inundation
60
40
20
0
0
20
40
60
Percent loss owing to climate change
80
APPLICATIONS III:
BIODIVERSITY LOSS IN MEXICO
Biodiversity Loss in Mexico
Loss of Evergreen Tropical Forest
Black – area lost; lightest gray – area remaining
Jay Species – Distributional Loss
Map of Distributional Loss - Corvids
APPLICATIONS IV: MALARIA
IN AFRICA
Anopheles gambiae
Forecasting Dengue Vector
Activity Using Time-specific
Ecological Niche Modeling
A. T. Peterson, Y. Nakazawa, and E. Martinez-Meyer
Instituto de Biologia and Facultad de Ciencias, Universidad
Nacional Autonoma de Mexico
Satellite Imagery
Advanced Very High Resolution Radiometer
(AVHRR) passes over all points on Earth’s
surface each day
Raw data include reflectance in different parts of
the electromagnetic spectrum
Normalized Difference Vegetation Index (NDVI)
is a composite of two bands of raw data that
summarizes ‘greenness’, and (more precisely)
volume of photosynthetic vegetation
AVHRR NDVI data are available as biweekly or
monthly composites
Potential for ENM and Dengue
ENM is presently developed based on static
environmental data that are not temporally precise
Using occurrence information and ecological data
that are precise both in time and space, ENM can
be made time-specific
Would provide detailed predictivity on a weekly or
monthly basis for vector distributions and disease
transmission
Potentially applicable to any ephemeral species of
epidemiological interest
Distributional Data for Aedes
aegypti, 1995, in Mexico
Time-ignorant Prediction
Closer View
Note broad areas predicted
Time-specific Occurrence Data
August 1995
August 1995 Predicts August 1995
(easy)
Can We Predict August from
Previous Months?
Example: June 1995 predicts August 1995
Composite Prediction for August 1995:
Two Previous Months Averaged
Time-specific Model Compared
with Time-ignorant model
Closer View
Summary of Prototype Tests
Percent correctly predicted
Month
N
Any
>50%
>80%
June
22
100
72.7
54.5
July
28
100
82.1
67.9
August
40
100
75
35
September
25
100
80
16
October
19
94.7
78.9
21.1
November
25
100
76
52
December
22
100
95.5
81.8
APPLICATIONS: EMERGENCE
OF LEISHMANIASIS IN
SOUTHERN BRAZIL
Diferencia 1975 – 2055 - migonei
Red areas improve for the species, blue areas get worse for the species
Diferencia 1975 – 2055 - intermedia
Red areas improve for the species, blue areas get worse for the species
Diferencia 1975 – 2055 - whitmani
Conclusion: Lutzomyia whitmani
is the sandfly species most likely to
be responding to climate change and
causing the observed re-emergence
Red areas improve for the species, blue areas get worse for the species
Niche Modeling Summary
Niches evolve slowly
Conserved niches allow many fascinating
insights
Characterize ecology and distributions
Predict unknown and undescribed populations
Predict geography of species’ invasions
Predict shifts in species’ distributions in
response to climate change and land use
change
Detect effects of interactions with other species
Improve conservation planning based on
species’ distributions
PLEASE, now
assemble, organize, and prepare
your good and bad vibrations
Thanks
[email protected]