OUTLINE. Distribution models. â presence-only, presence-background, presence-absence. Case studies. Fuzzy logic. Model evaluation. Take-home messages ...
Is there safety in numbers? Can we be confident with our species distribution models? A. Márcia Barbosa CIBIO/InBIO – UÉvora (Portugal)
OUTLINE Distribution models → presence-only, presence-background, presence-absence
Case studies
Fuzzy logic Model evaluation
Take-home messages
Distribution models species distribution data geology temperature
precipitation
environmental variables
topography
species’ predicted distribution
Presence-only methods
geographical envelopes
(kernels, hulls – range maps)
environmental envelopes (OA, BIOCLIM, DOMAIN)
Presence-background methods
compare distributions of the variables between the presences and the available area (ENFA, GARP, MAXENT)
Presence-absence methods
GLM, GAM, CART, BRT, RF, ANN, MARS
- good transparency/complexity balance - several success stories:
Baetic midwife toad (Alytes dickhilleni) Herp conference 2004
Baetic midwife toad (Alytes dickhilleni) Herp conference 2004 Herp conference 2012
- 38 new presences (+28%) - mean favourability 0.87
- proportion of new records in favourable (>0.5 or >0.8) sites higher than chance - correct classification and kappa increased
Monk parakeet (Myiopsitta monachus)
Distribution in 2003
Predicted favourability
Muñoz & Real (2006) Diversity and Distributions 12: 656-665
Monk parakeet (Myiopsitta monachus)
Distribution in 2003
Predicted favourability
Muñoz & Real (2006) Diversity and Distributions 12: 656-665
Monk parakeet (Myiopsitta monachus)
2003: intermediate fav.
high favourability presence data 2003
2015: presence
Monk parakeet (Myiopsitta monachus)
2003: intermediate fav.
high favourability presence data 2003
2015: presence
Iberian painted frogs (Discoglossus sp.)
D. galganoi D. jeanneae Real et al. (2005) Canadian Journal of Zoology 83: 536-545
Real et al. (2005) Canadian Journal of Zoology 83: 536-545
Iberian lynx (Lynx pardinus)
Real et al. (2009) Diversity and Distributions 15: 390-400
Iberian lynx (Lynx pardinus) correlated with abundance
Favourability for lynx+rabbit
max
min Caribú Hongo
Real et al. (2009) Diversity and Distributions 15: 390-400
Iberian desman (Galemys pyrenaicus)
Barbosa et al. (2010) Ecological Modelling 220:747-754
conservation? Barbosa et al. (2010) Ecological Modelling 220:747-754
0.00 0.06 0.13 0.19 0.25 0.31 0.38 0.44 0.50 0.56 0.63 0.69 0.75 0.81 0.88 0.94 1.00
0 1
Barbosa et al. (2010) Conservation Biology 24: 1378-1387
validation? Barbosa et al. (2010) Conservation Biology 24: 1378-1387
1
Validation: Favourability at sampled vs. random points (1000 samples / region) Favourability at presences vs. absences
Proportion presences in favourable / unfavourable vs. random areas Favourability at SICs vs. remaining territory
Barbosa et al. (2010) Conservation Biology 24: 1378-1387
0
Eurasian otter (Lutra lutra)
Barbosa et al. (2010) Biological Conservation 114:377-387
Eurasian otter (Lutra lutra)
Barbosa et al. (2010) Biological Conservation 114:377-387
Barbosa et al. (2010) Conservation Biology 24: 1378-1387
Iberian ribbed newt (Pleurodeles waltl)
Iberian ribbed newt (Pleurodeles waltl)
Last Interglacial (LIG)
Last Glacial Maximum (LGM)
Present
models identify historical reservoirs for genetic diversity
Intersection of favourability in LIG-LGM
PCA axis 1 of 4 genetic diversity variables
Species distribution models and fuzzy logic
Microtus cabrerae M. guentheri
M. thomasi
Species distribution models and fuzzy logic
M. guentheri
M. thomasi
Species distribution models and fuzzy logic
incorporates
uncertainty
Barbosa (2015) Methods in Ecology and Evolution, in press
Species distribution models and fuzzy logic
⇨ combining models ⇨ biogeographic regions ⇨ beta diversity (loss, gain, turnover)
Barbosa (2015) Methods in Ecology and Evolution, in press
70% 20%
Model evaluation
70% 20%
- discrimination / classification - calibration / reliability
Things to keep in mind…
"[model] values reflect the probability that each cell is occupied by the focal species, based on the covariate values associated with the cell. These
probabilities
can
be
viewed
as
expectations
associated with a stochastic process and do not represent realizations of this process.“ (Karanth et al. 2009, J. Appl. Ecol.)
Things to keep in mind… "Under the most rigorously controlled conditions of pressure, temperature, volume, humidity, and other
variables, an organism will do as it damn well pleases.“ (Murphy’s Law)
"Essentially, all models are wrong, but some are useful."
(George E. P. Box)
Is there safety in numbers?
Can we be confident with our species distribution models?
rubbish in, rubbish out!
that’s it, thanks!
fuzzy sets and fuzzy logic
Barbosa (2015) Methods in Ecology and Evolution, in press
fuzzy sets and fuzzy logic
Barbosa (2015) Methods in Ecology and Evolution, in press