Ecological

1
Ecological Niche Modelling
2
Ecological Niche models

• Empirical models that predict the spatial
  distribution of a species from the environmental
  conditions at the locations where it is known to
  be present (or absent).
• Assumption The environment at the locations
  where a species occurs is represents its
  ecological niche.
• A plethora of statistical approaches is used

3
Niche Modeling

• Get occurrence points
• Extract environmental data
• Build statistical model
• Predict
• Evaluate
• Use

4
Batrachoseps regius
Batrachoseps luciae
5
Presence/Absence
6
3. Formulate model
Presence only BIOCLIM, DOMAIN Presence /
Absence (Background) Machine Learning ANN,
CART, MAXENT, SVM, Boosting
Regression Logisitic, GLM, GAM, MARS
7
BIOCLIM (Presence only)

• Calculate percentile distributions for each
  environmental variable
• If p gt 0.5 then p 1 - p
• For each site
• For each env. var
• calculate place in distribution
• keep lowest score acrossenv vars.

8
0-100 percentile
BIOCLIM
2.5 -97.5 percentile
Envelope model
5 - 95 percentile
9
4. Predict
BIOCLIM B. lucia
10
B. regius
11
DOMAIN (Presence only)
Gower distance between occurrence A and site B
k environmental variable
12
B. lucia
13
DOMAIN
BIOCLIM
14
B. regius
15
B. regius
DOMAIN
BIOCLIM
16
Classification and Regression Trees (Machine
learning Presence / Absence)
A - Temperature seasonality (STD) B - Monthly
Temperature range C - Precipitation seasonality
(CV) D - Annual precipitation
false A lt 2.86 true f C gt 96.1
t f B gt 11.3 t 0 f D lt
419 t 0 f A lt 2.27 t
0 0.1 0.4
0.8
17
CART B. lucia
18
MAXENT (Presence / Background)

• Machine learning technique
• Assume maximum entropy (i.e. similar to
  uniform distribution) But fit empirical data
• Maximum likelihood Gibbs distribution
  (exponential in a linear combination of the
  environmental variables)

19
MAXENT B. lucia
20
MAXENT B. regius
21
Regression (Presence / Absence)
Least square regression General Linear Models
(non Gaussian) General Additive Models (non
parametric)
22
Select variables (GRASP)
23
Responses
Annual Temperature
Monthly Temperature range
Temperature seasonality
Precipitation ofWarmest quarter
Precipitation Seasonality
Annual Precipitation
24
(No Transcript)
25
Issues (applied)

• Model choice
• Environmental data selection
• Threshold selection
• Data quality / Bias