Transcript Habitat Modeling - Central Michigan University

Habitat Modeling
Goals

Predict the locations of as-yet
undiscovered refuges in the Great Lakes

Develop management protocols to
create new unionid habitat
Goals

Predict the locations of as-yet
undiscovered refuges in the Great
Lakes
 what habitat parameters are necessary to
sustain unionid populations
 develop a GIS-based model that will
summarize all the important features of the
refuges.
◦ Test models predictions
◦ Use an iterative process to refine the
model.

Habitat parameters important for
unionid protection from zebra
mussels may include:
◦ presence of substrates soft enough for
unionids to burrow into
◦ large areas of shallow waters (protected
bayous) with low flow and warmer
temperatures that encourage unionid
burrowing
◦ hydrological connection of the bayous
to the lake
◦ fish predation of Dreissena attached to
unionids
◦ Interactions of all these factors.

Factors that inhibit the establishment of
stable dreissenid populations are:
◦ wave action in shallow areas, water level
fluctuations, ice scouring
◦ dense reed-beds
◦ remoteness from the source of dreissenid
veligers
◦ In addition, there may be other, yet
unidentified, mechanisms that promote the
long-term coexistence of dreissenids and
native mussels.
At the local scale

Focus on areas inhabited by mussels:
◦ substrate type,
◦ depths,
◦ water temperature,
◦ water velocity
◦ location
◦ species richness and abundance.

Use multivariate methods such as multiscaled
ordination with CCA (MSO-CCA) to define local
scale habitat.
At a regional scale

Use ecological niche modeling to predict the
potential presence or absence of mussel beds.

Lots of options for model types, GARP, SVM, CART,
etc.

Use available environmental data
◦ water depth, wind-driven currents, mean, maximum and
minimum annual temperature.

Developed GIS data layers
◦ Turbidity, distance to deep-water, bay area and shape,
bottom oxygen, distance to rivers, and human-related
factors, such as distance to nearest dredging operation
and distance to dams in upstream rivers.
Ecological Niche Model

Predicted the potential distribution of zebra mussels.


Based on current distribution of zebra mussels in U.S.
11 geologic and environmental variables.

Biological model - 6 factors that have plausible explanations
for limiting the distribution of zebra mussels.


frost frequency, maximum annual temperature, elevation, slope,
bedrock geology, and surface geology.
No Elevation model
Drake & Bossenbroek, 2004, Bioscience
Biological Model
Biological Mode
Predicted
Value
Distribution
100%
¯
0%
Zebra Mussel
ZM
Locations
Kilometers
0
250 500
1,000
Biological Model minus Elevation
No Elevation
Predicted
Value
Distribution
High : 100
¯
Low : 0
Zebra Mussel
ZM
Locations
Kilometers
0
250 500
1,000
Support Vector Data Description
The support vector data description
(SVDD) is an SVM for finding the
boundary around a set of observations.
 This boundary is the simplest boundary in
the sense that it represents the smallest
possible hyper- volume (a hypersphere)
containing a specified fraction of the
observations in the projected feature
space

Support Vector Data Description
Drake & Bossenbroek, 2009, Theor. Ecol.
Questions?