Successfully reported this slideshow.
We use your LinkedIn profile and activity data to personalize ads and to show you more relevant ads. You can change your ad preferences anytime.
Observability and Multi-species Range Models Bob O'Hara BiK-F Frankfurt am Main, Germany
Want to Know What affects Species Distributions Source: mrpink http://www.flickr.com/photos/mjadamo/28640174/
Cottage Industry: Niche Models
Presence Only Models
Dealing With Presence Only Data Create “pseudo-absences” From all of the places a species is not found,  randomly select s...
Problem: False Absences Sampling  effort varies Pseudo-absence  methods deal with  this poorly WE NEED SOMETHING BETTER
Estimating Sampling Effort Repeated visits:  can use absences Use reports of other species as estimate of effort?
Data Database of UK butterfly records Use 8 most common species 10km x 10km squares Covariates: proportion of habitat
The Species Images from www.flickr.com Green Veined White  Pieris napi Source: oldbilluk  /oldbilluk/3504529745/ Red Admir...
The  Data  Square root of  no. of visits shown
The Model Process Model and  Sampling Model http://pixdaus.com/single.php?id=235258&from=email
Process Model I i  – presence of species  s  at site  i X ij  – proportion of habitat of type  j  at location  i G ij  – L...
SSVS P( b ) = Pr( I =0) N(0,  s b 2 ) + Pr( I =1) N(0,  c   s b 2 ) Mixture of Normals c=1000 Spike Slab
Observation Model Y i ( s )– Species  s  observed in sample  j Pr ( Y i ( s ) = 1 |  I i  = 0) = 0 logit( Pr ( Y i ( s ) =...
Implementation MCMC: OpenBUGS Vague Priors 2 chains, 10k iterations Large data set -> takes a few days
The Results
Maps Black: P=1
Small White
Results: Observation Effort Darker= more effort
Habitat Effects Black=positive, red=negative
Species' Detectabilities
Lat  and  Long  Effects 50% Confidence regions
So? We Can Do It Should improve the predictions Uses the data more efficiently
For the Future Add climate Multi-species interactions at process level? http://services.niagaracollege.ca/unitedway/virtua...
Upcoming SlideShare
Loading in …5
×

Multispecies Distribution Models

788 views

Published on

Slides for talk at ISEC, 2010 (University of Canterbury, Kent, UK).

Published in: Education, Technology
  • Be the first to comment

  • Be the first to like this

Multispecies Distribution Models

  1. 1. Observability and Multi-species Range Models Bob O'Hara BiK-F Frankfurt am Main, Germany
  2. 2. Want to Know What affects Species Distributions Source: mrpink http://www.flickr.com/photos/mjadamo/28640174/
  3. 3. Cottage Industry: Niche Models
  4. 4. Presence Only Models
  5. 5. Dealing With Presence Only Data Create “pseudo-absences” From all of the places a species is not found, randomly select some to act as absences
  6. 6. Problem: False Absences Sampling effort varies Pseudo-absence methods deal with this poorly WE NEED SOMETHING BETTER
  7. 7. Estimating Sampling Effort Repeated visits: can use absences Use reports of other species as estimate of effort?
  8. 8. Data Database of UK butterfly records Use 8 most common species 10km x 10km squares Covariates: proportion of habitat
  9. 9. The Species Images from www.flickr.com Green Veined White Pieris napi Source: oldbilluk /oldbilluk/3504529745/ Red Admiral Vanessa atalanta Source: Durlston Country Park durlston/2891053154/ Large White Pieris brassicae Source: jpockele /jpockele/196916845/ Meadow Brown Maniola jurtina Source: Durlston Country Park durlston/2553616249/ Speckled Wood Pararge aegeria Source: Yersinia yersinia/3465615645/ Small Tortoiseshell Aglais urticae Source: Wanja Krah wanjakrah/4552290666/ Peacock Inachis io Source: Chris@184 chrisbradbury/2745513954/ Small White Pieris rapae Source: MarcelGermain marcelgermain/1935857010/
  10. 10. The Data Square root of no. of visits shown
  11. 11. The Model Process Model and Sampling Model http://pixdaus.com/single.php?id=235258&from=email
  12. 12. Process Model I i – presence of species s at site i X ij – proportion of habitat of type j at location i G ij – Latitude/longitude at location i
  13. 13. SSVS P( b ) = Pr( I =0) N(0, s b 2 ) + Pr( I =1) N(0, c s b 2 ) Mixture of Normals c=1000 Spike Slab
  14. 14. Observation Model Y i ( s )– Species s observed in sample j Pr ( Y i ( s ) = 1 | I i = 0) = 0 logit( Pr ( Y i ( s ) = 1 | I i = 1)) = f ( s ) + y i Observed|Present = Site + Species
  15. 15. Implementation MCMC: OpenBUGS Vague Priors 2 chains, 10k iterations Large data set -> takes a few days
  16. 16. The Results
  17. 17. Maps Black: P=1
  18. 18. Small White
  19. 19. Results: Observation Effort Darker= more effort
  20. 20. Habitat Effects Black=positive, red=negative
  21. 21. Species' Detectabilities
  22. 22. Lat and Long Effects 50% Confidence regions
  23. 23. So? We Can Do It Should improve the predictions Uses the data more efficiently
  24. 24. For the Future Add climate Multi-species interactions at process level? http://services.niagaracollege.ca/unitedway/virtual%20break.htm

×