Cindy Kolar

1
Biological risk assessment modeling for
potentially invasive species
Cindy Kolar Upper Midwest Environmental Sciences
Center Risk Assessment Workshop August, 2005
U.S. Department of the Interior U.S. Geological
Survey
Upper Midwest Environmental Sciences Center
2
Talk Outline
1. Why look at biological characteristics?
2. Qualitative methods
3. Quantitative methods
4. Developing and using quantitative models for
fishes in Great Lakes
5. Exercise using models for Great Lakes
Upper Midwest Environmental Sciences Center
3
Hypothesized Characteristics of IS

• Broad diet
• Single-parent reproduction
• High genetic variability
• Phenotypically plastic

• Large native range
• Gregarious
• Long-lived
• Human commensal

• Strategy (r-selection or switch between k- and
  r-?)
• Individual size (small or large?)
• Population density (constant or boom and bust?)

(from Lodge 1993)
Upper Midwest Environmental Sciences Center
4
Productive Approaches for Biological Risk
Assessments

• Transition step-specific controls for inter-step
  differences
• Region-specific controls for species-ecosystem
  interaction
• Taxon-specific controls for inter-taxa
  differences

Upper Midwest Environmental Sciences Center
5
Selected Invasion-Associated Characteristics
PLANTS
BIRDS
Invasive/Not
Establish/Fail
Invasive/Not
Establish/Fail
Characters
(modified from Kolar Lodge, 2001 TREE
16199-204)
Upper Midwest Environmental Sciences Center
6
Selected Invasion-Associated Characteristics
PLANTS
BIRDS
Invasive/Not
Establish/Fail
Invasive/Not
Establish/Fail
Characters
Body mass
, , ns, ns, ns
-
-, ns, ns, ns, ns

Migrating

ns, ns
Len. flowering
(modified from Kolar Lodge, 2001 TREE
16199-204)
Upper Midwest Environmental Sciences Center
7
Selected Invasion-Associated Characteristics
PLANTS
BIRDS
Invasive/Not
Establish/Fail
Invasive/Not
Establish/Fail
Characters
Body mass
, , ns, ns, ns
-
-, ns, ns, ns, ns

Migrating

ns, ns
Len. flowering
Invasion history
, , ,
, , ,
Family invasive
, ,
Vegetative repro.
(modified from Kolar Lodge, 2001 TREE
16199-204)
Upper Midwest Environmental Sciences Center
8
Selected Invasion-Associated Characteristics
PLANTS
BIRDS
Invasive/Not
Establish/Fail
Invasive/Not
Establish/Fail
Characters
Body mass
, , ns, ns, ns
-
-, ns, ns, ns, ns

Migrating

ns, ns
Len. flowering
Invasion history
, , ,
, , ,
Family invasive
, ,
Vegetative repro.
Annual (vs. per.)
ns, ns, ns
ns, ns
Diet breadth
ns
ns
Diverse climates
(modified from Kolar Lodge, 2001 TREE
16199-204)
Upper Midwest Environmental Sciences Center
9
Using Species Characteristics

• Patterns emerging from focused studies (one
  taxon, one ecosystem, one region)

• Some species ecosystem characteristics have
  consistent association with invaders

• But ability to predict using species
  characteristics limited by small sample sizes,
  a lack of diversity of taxa studied

Upper Midwest Environmental Sciences Center
10
Talk Outline
1. Why look at biological characteristics?
2. Qualitative methods
3. Quantitative methods
4. Developing and using quantitative models for
fishes in Great Lakes
5. Exercise using models for Great Lakes
Upper Midwest Environmental Sciences Center
11
Qualitative Modeling of Biological Risk

• ANSTF Generic Risk Assessment

• Scoring systems such as the Weed Risk
• Assessment (Australia)

• Common characteristics of invasive species

Upper Midwest Environmental Sciences Center
12
Qualitative Modeling of Biological Risk
Ricciardi and Rasmussen (1998) common sense
approach
Temperature tolerance
Yes
No
Fail
Invasion history
No
Yes
Fail
Propagule pressure
High
Low
Low success
High success
Upper Midwest Environmental Sciences Center
13
Talk Outline
1. Why look at biological characteristics?
2. Qualitative methods
3. Quantitative methods
4. Developing and using quantitative models for
fishes in Great Lakes
5. Exercise using models for Great Lakes
Upper Midwest Environmental Sciences Center
14
Quantitative Modeling of Biological Risk
Multivariate Statistical Modeling
1. Discriminant Analysis (DA) two or more
groups know membership
2. Canonical Discriminant Analysis (CDA) uses
new variables
3. Logistic Regression (LR) two groups
normality not necessary
4. Cluster Analysis (CA) classifies when dont
know membership
Upper Midwest Environmental Sciences Center
15
Quantitative Modeling of Biological Risk
Decision Trees Categorical Regression Trees
(CART)
Developed by Breiman et al. (1984)
Upper Midwest Environmental Sciences Center
16
Rule Sets are Key Elements CART Analyses
1. Splitting each node in tree
Considers all possible splits for EACH variable
Ranks by a quality-of-split criterion and splits
on top ranked variable
2. Deciding when tree is complete
Overgrows then prunes back
3. Assigning each terminal node to a class
outcome
Ex plurality rule--group with greatest
representation determines class assignment
4. Testing
Lots data build tree with learning sample, then
calculate misclassification rate using test sample
Less data bootstrap cross validation technique
Upper Midwest Environmental Sciences Center
17
Studies using CART in Biological Abstracts
35
Ecology/Applied
30
Health Field
25
20
Number of studies
15
10
5
0
1990-
1996-
1994-
1998-
1992-
1991
1997
1995
1999
1993
Upper Midwest Environmental Sciences Center
18
Comparison of Quantitative Methods
DA
CART
Two or more groups
X
X
X
X
Independent variables
Populations are distinct
X
X
Multivariate normality
X
Equal covariance matricies
X
Mathematical function
X
Decision tree
X
Common statistical packages
X
Stand alone software or expensive
X
Upper Midwest Environmental Sciences Center
19
Talk Outline
1. Why look at biological characteristics?
2. Qualitative methods
3. Quantitative methods
4. Developing and using quantitative models for
fishes in Great Lakes
5. Exercise using models for Great Lakes
Upper Midwest Environmental Sciences Center
20
Pathways for Introduction of Fishes
Sea lamprey
1.
Adjacent watersheds
2.
Stocking
Grass carp
3.
Ballast water
4.
Live bait trade
Brown trout
5.
Aquarium industry
6.
Aquaculture
Weather loach
Round goby
Rudd
Upper Midwest Environmental Sciences Center
21
Productive Approaches Quantitative Predictions

• Region-specific controls for species-ecosystem
  interaction
• Taxon-specific controls for inter-taxa
  differences
• Transition step-specific controls for inter-step
  differences

Upper Midwest Environmental Sciences Center
22
Process of Species Spread
Transportation

• Develop predictive
• models

Introduction

• Use models in
• risk assessment

Establishment
Spread
Impact
Slowly
Non- nuisance
Quickly
Nuisance
Upper Midwest Environmental Sciences Center
23
Habitat Environmental Tolerances
Area of native range (km2)
Minimum temperature threshold
Maximum temperature threshold
Range of temperature tolerances
Range salinity tolerance (scale freshwater only
to marine)
Human use (scale pest to important sport/comml
fishery)
Upper Midwest Environmental Sciences Center
24
Life History Characteristics
Length at maturity (mm)
Egg diameter (mm)
Age at maturity (yrs.)
Hatch length (mm)
Fecundity (annual)
First yr growth ( mature)
Second yr growth ( mature)
Maximum yrs. spawn
Diet breadth ( food types)
Reproductive potential
Incubation period (days)
Longevity (yrs.)
Parental care (scale 1-7)
Degree of derived characters
Upper Midwest Environmental Sciences Center
25
Additional Variables
Whether the genus has history of introduction,
establishment, or invasiveness elsewhere
Whether the species has history of
introduction, establishment, or invasiveness
elsewhere
Upper Midwest Environmental Sciences Center
26
Database Complete
Upper Midwest Environmental Sciences Center
27
Question
1. Are fishes that successfully invaded the
Great Lakes different from those that have
failed?
Upper Midwest Environmental Sciences Center
28
Establishment of Fishes
Failed (n 21)
Successful (n 24)
rainbow smelt
Atlantic salmon
Upper Midwest Environmental Sciences Center
29
Discriminant Function

• Fast relative growth rate
• Wide salinity tolerance
• Wide range of water temperature tolerance
• Species has a history of invasiveness

Overall Correct Classification 87
Upper Midwest Environmental Sciences Center
30
Question
2. Are fishes that quickly spread through the
Great Lakes different from those that spread
slowly?
Establishment
Quickly spreading (n9)
Spread
2
Round goby
Slowly
Quickly
Slowly spreading (n8)
Overall correct classification 89
Shortnose gar
Upper Midwest Environmental Sciences Center
31
Questions
3. Are fishes that are perceived as a nuisance
in the Great Lakes different from those that
are not?
Nuisance (n8)
Establishment
Impact
3
Common carp
Non- nuisance
Nuisance
Non-nuisance (n15)
Oriental weatherfish
Overall correct classification 91
Upper Midwest Environmental Sciences Center
32
CART Decision Tree for Establishment
Relative growth by 2 yrs
68.5

Number taxa in diet
Minimum temperature
4.5

5.5

FAIL
SUCCESS
Number taxa in diet
Relative growth by 1 yr
0 Success 5 Fail
3 Success 1 Fail
1.5


26.5
SUCCESS
FAIL
1 Success 0 Fail
0 Success 13 Fail
FAIL
SUCCESS
0 Success 1 Fail
20 Success 1 Fail
Overall correct classification rate 96
Upper Midwest Environmental Sciences Center
33
Predictive Models Developed
1. Establishment

• Discriminant function

• CART tree

2. Spread

• Discriminant function

3. Impact

• Discriminant function

Upper Midwest Environmental Sciences Center
34
Application Ponto Caspian fishes
Sea of Azov
Black Sea
Caspian Sea
66 out of 110 species
Mediterranean Sea
Upper Midwest Environmental Sciences Center
35
Establishment of Ponto-Caspian Fishes
Upper Midwest Environmental Sciences Center
36
Quick Spreading Nuisance Fishes
22 species common to both models
Spread
Impact
Slowly
Quickly
Non- nuisance
Nuisance
Upper Midwest Environmental Sciences Center
37
Highest Risk Ponto-Caspian Fishes
Tyulka (Clupeonella cultriventris)
Monkey goby (Neogobius fluviatilis)
Eurasian minnow (Phoxinus phoxinus)
Black sea silverside (Atherina boyeri)
European perch (Perca fluviatilis)
Upper Midwest Environmental Sciences Center
38
Uses of Biological Risk Models
Intentional Pathways of Introduction

• Basis for developing regulations

• Basis for developing best management practices

• Basis for developing introduction policies

Upper Midwest Environmental Sciences Center
39
Statistical Models Decision Trees
Upper Midwest Environmental Sciences Center
40
Talk Outline
1. Why look at biological characteristics?
2. Qualitative methods
3. Quantitative methods
4. Developing and using quantitative models for
fishes in Great Lakes
5. Exercise using models for Great Lakes
Upper Midwest Environmental Sciences Center
41
Combination Decision Tree
Can it survive in freshwater water temperatures

YES
NO
Does it mature by 2 years?
FAIL
NO
YES
Is it invasive elsewhere?
Is it invasive elsewhere?
YES
NO
Can it survive in brackish water?
Is it 85 adult length by 1 yr?
YES
NO
SUCCEED
YES
NO
YES
NO
Can it survive in brackish water?
SUCCEED
SUCCEED
FAIL
Is it 68 adult length by 2 yrs?
YES
NO
YES
NO
SUCCEED
FAIL
FAIL
SUCCEED
Overall correct classification rate 91
Upper Midwest Environmental Sciences Center
42
Evaluation of Analyses
Jack-knife classification rates
False -
Correct
Correct -
False
Question
Sp. characteristics
1. Establishment
75
91
25
9
DA
Rel. growth (), Salinity tol. (), Range temp.
tol. (), Sp. history invas.()
96
71
4
29
CART
Rel. growth (), Diet breadth (), Min. temp. (-)
78
100
22
0
Max. temp. tol. (-), Rel. growth(), Range temp.
tol. ()
2. Spread
DA
3. Impact
92
90
8
10
Egg diameter(-), Min. temp. tol. (-), Salinity
tol. ()
DA