Working with what we know – presence-only / ecological niche models in marine mammal science Kristin Kaschner, Colin MacLeod, Laura Mandleberg, Ross Comptom Sea Around Us Project, Fisheries Centre, University of British Columbia, Canada & FTZ Büsum, Kiel Christian-Albrechts-University, Germany School of Biological Sciences (Zoology), University of Aberdeen, UK
66
Embed
Working with what we know – presence-only / ecological ... · WHAT? What are they? • Predict ecological niches • Use only presence data
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Working with what we know –presence-only / ecological niche models
in marine mammal science
Kristin Kaschner, Colin MacLeod, Laura Mandleberg, Ross Comptom
Sea Around Us Project, Fisheries Centre, University of British Columbia, Canada &FTZ Büsum, Kiel Christian-Albrechts-University, Germany
School of Biological Sciences (Zoology), University of Aberdeen, UK
Outline
• What are presence-only models?
• Why do we use them?
• Which ones are there?
• Do they work?
• What can we do with them?
OUTLINE
What are they?WHAT?
• Predict ecological niches
• Use only presence data
Why do we use them?WHY?
• Data paucity
• Absence data issues
• Niche modeling vs. distribution
Data paucity
Hawaii
WHY?
Data paucityWHY?
OBIS-SEAMAP (http://seamap.env.duke.edu/species)
- compilation & storage of marine mammal occurrence data
- out of 115 species, geo-referenced / effort corrected data
- available/accessible for ~ 50%
- representative coverage = ~ 2 %
Data paucityWHY?
Model evaluationPresence-absence confusion matrix
Predicted presence
Predicted absence
Recorded presence
Recorded absence
a (true presence)
c (false absence)
b (false presence)
d (true absence)
WHY? Absence Data Issues
Model evaluationPresence-absence confusion matrix
Predicted presence
Predicted absence
Recorded presence
Recorded absence
a (true presence)
c (false absence)
b (false presence)
d (true absence)
WHY? Absence Data Issues
Omission error / Model overfitting
Commission error / Model overprediction
Model evaluationPresence-absence confusion matrix
Predicted presence
Predicted absence
Recorded presence
Recorded absence
a (true presence)
c (false absence)
b (false presence)
d (true or perceived absence????)
WHY? Absence Data Issues
Site with environmental
value X
Species present?
Site visited? Species detected?
Presence-only data
M. Nakamura, CONABIO, 2005GBIF Ecological Niche Modelling Workshop, KU
Absence Data IssuesWHY?
Site with environmental
value X
Species present?
Site visited? Species detected?
Presence-absence data
M. Nakamura, CONABIO, 2005GBIF Ecological Niche Modelling Workshop, KU
Absence Data IssuesWHY?
Site with environmental
value X
Species present?
Site visited? Species detected?
True absence data
M. Nakamura, CONABIO, 2005GBIF Ecological Niche Modelling Workshop, KU
Absence Data Issues
False absence data
WHY?
Absence Data IssuesWHY?
0
0.5
1
1.5
2
2.5
0 20 40 60 80Depth [m]
Mea
n de
nsity
[# a
nim
als
/ km
2]
Harbour porpoise density
No animals in deeper waters!
Scheidat, Gilles et al, (unpublished data)
Absence Data IssuesWHY?
0
0.5
1
1.5
2
2.5
0 20 40 60 80Depth [m]
Mea
n de
nsity
[# a
nim
als
/ km
2]
0
2
4
6
8
10
12
0 20 40 60 80Depth [m]
Mea
n Ef
fort
[km
2]
Harbour porpoise density
Aerial survey effort
No animals in deeper waters?
True absences???
Scheidat, Gilles et al, (unpublished data)
0
10
20
30
40
50
60
0 5 10 15 20 25
Effort [km2]
# si
ghtin
gs
Absence Data IssuesWHY?
Spring 2002-2005
Scheidat, Gilles et al, (unpublished data)
Absence Data IssuesWHY?
Spring
0
10
20
30
40
50
60
70
80
0 10 20 30Effort category [km2]
Mea
n #
sigh
tings
Summer
0
10
20
30
40
50
60
70
80
0 10 20 30Effort category [km2]
Mea
n #
sigh
tings
Fall
0
10
20
30
40
50
60
70
80
0 10 20 30Effort category [km2]
Mea
n #
sigh
tings
Scheidat, Gilles et al, (unpublished data)
Absence Data IssuesWHY?
Spring
R2 = 0.4597
0
10
20
30
40
50
60
70
80
0 10 20 30Effort category [km2]
Mea
n #
sigh
tings
Summer
R2 = 0.6622
0
10
20
30
40
50
60
70
80
0 10 20 30Effort category [km2]
Mea
n #
sigh
tings
Fall
R2 = 0.6069
0
10
20
30
40
50
60
70
80
0 10 20 30Effort category [km2]
Mea
n #
sigh
tings
Scheidat, Gilles et al, (unpublished data)
Absence Data IssuesWHY?
Spring
0
10
20
30
40
50
60
70
80
0 10 20 30Effort category [km2]
Mea
n #
sigh
tings
Summer
0
10
20
30
40
50
60
70
80
0 10 20 30Effort category [km2]
Mea
n #
sigh
tings
Fall
0
10
20
30
40
50
60
70
80
0 10 20 30Effort category [km2]
Mea
n #
sigh
tings
True absences???
Scheidat, Gilles et al, (unpublished data)
.20
Variable A
Species #1
Presence
.8Bias
.16Obs. rate
Example: 100 trials
16Observed
.80
Variable B
.1
.08
8
0
Variable C
.1
0
0
× ××
= ==
.32
Variable A
Species #2
.5
.16
16
.16
Variable B
.5
.08
8
.52
Variable C
0
0
0
×× ×
== =
M. Nakamura, CONABIO, 2005GBIF Ecological Niche Modelling Workshop, KU
Absence Data IssuesWHY?
Model Complexity
Pre
dict
ion
Err
or
Low High
Low Variance High Variance
Training sample
Test sample
Hastie et al. (2001)
WHY? Ecological Niche vs Distribution
Model Complexity
Pre
dict
ion
Err
or
Low High
Low Variance High Variance
Training sample
Test sample
Hastie et al. (2001)
WHY? Ecological Niche vs Distribution
Ecological Niche Model
Distribution Model
WHY? Ecological Niche vs Distribution
J. Soberon, CONABIO, 2005GBIF Ecological Niche Modelling Workshop, KU