Gf o2014talk
0 likes724 views
This document presents a unified framework for combining different data types in species distribution modelling. It proposes using a single continuous density model rather than separate models. Observation data like presence-only points, abundance counts, and expert ranges can then be modeled as coming from the same underlying density through different observation processes. This allows integrating data from various sources like museums, citizen science, and surveys in a unified Bayesian modeling approach using integrated nested Laplace approximations for efficient inference. The framework is demonstrated on modeling the distribution of the solitary tinamou using different data types.
Gf o2014talk
- 1. A unified framework to combine disperate data types in species distribution modelling A unified framework to combine disperate data types in species distribution modelling Slides on Slideshare: http://www.slideshare.net/oharar/gf-o2014talk Bob O’Hara1 Petr Keil 2 Walter Jetz2 1BiK-F, Biodiversity and Climate Change Research Centre Frankfurt am Main Germany Twitter: @bobohara 2Department of Ecology and Evolutionary Biology Yale University New Haven, CT, USA
- 2. A unified framework to combine disperate data types in species distribution modelling A ”Real”Curve 0 20 40 60 80 100 020406080 Curve
- 3. A unified framework to combine disperate data types in species distribution modelling Approximated with a Discretised Curve 0 20 40 60 80 100 020406080 Curve Discrete
- 4. A unified framework to combine disperate data types in species distribution modelling Better: linear interpolation 0 20 40 60 80 100 020406080 Curve Discrete Interpolated
- 5. A unified framework to combine disperate data types in species distribution modelling With more points, the approximations improve 0 20 40 60 80 100 020406080 Curve Discrete Interpolated
- 6. A unified framework to combine disperate data types in species distribution modelling What does this have to do with distribution models?
- 7. A unified framework to combine disperate data types in species distribution modelling What does this have to do with distribution models? This is how SDMs see the world: source: http://bit.ly/1l8sG7M Map produced by Peter Blancher, Science and Technology Branch, Environment Canada, based on data from the North American Breeding Bird Survey
- 8. A unified framework to combine disperate data types in species distribution modelling Problems: scale, within-grid heterogeneity
- 9. A unified framework to combine disperate data types in species distribution modelling Let’s sidestep the whole problem Work in continuous space instead The maths will let us work on different scales e.g. Renner & Warton (2013) doi: 10.1111/j.1541-0420.2012.01824.x Lets us deal with points & irregular shapes Makes it straightforward to include different sorts of data
- 10. A unified framework to combine disperate data types in species distribution modelling Motivation Map Of Life www.mol.org/ Different data sources GBIF expert range maps eBird and similar citizen science efforts organised surveys (BBS, BMSs) Regional checklists
- 11. A unified framework to combine disperate data types in species distribution modelling A Unified Model There is a single state - density of the species Actual State Presence Absence Presence Only Expert Range Maps ¨ ¨¨% c r rrj
- 12. A unified framework to combine disperate data types in species distribution modelling Point Processes: Model Each point in space, ξ, has an intensity, ρ(ξ) log(ρ(ξ)) = η(ξ) = βX(ξ)+ν(ξ) The number of individuals in an area A follows a Poisson distibution with mean λ(A) = A ρ(ξ)ds
- 13. A unified framework to combine disperate data types in species distribution modelling Point Processes: Reality Approximate λ(ξ) numerically: select some integration points, and sum over those λ(A) ≈ N s=1 |A(s)|eη(s)
- 14. A unified framework to combine disperate data types in species distribution modelling Observation Models Presence only points: thinned point process Abundance: Poisson Presence/Absence: binomial, cloglog with µA(A, t) = η(A) + log(|A|) + log(t) + log(p) (large) areas: Pr(n(A) > 0) = 1 − e A eρ(ξ)dξ Expert range: use distance to range as a covariate
- 15. A unified framework to combine disperate data types in species distribution modelling Put these together Data likelihoods: P(Xi |λ) for data Xi . Total likelihood is P(X) = i P(Xi |λ)P(λ) Where P(λ) is the actual distribution model, and will depend on environmental and other covariates
- 16. A unified framework to combine disperate data types in species distribution modelling In practice Be Bayesian. Could use MCMC, but this is quicker in INLA SolTim.res <- inla(SolTim.formula, family=c('poisson','binomial'), data=inla.stack.data(stk.all), control.family = list(list(link = "log"), list(link = "cloglog")), control.predictor=list(A=inla.stack.A(stk.all)), Ntrials=1, E=inla.stack.data(stk.all)$e, verbose=FALSE)
- 17. A unified framework to combine disperate data types in species distribution modelling The Solitary Tinamou Photo credit: Francesco Veronesi on Flickr (https://www.flickr.com/photos/francesco veronesi/12797666343)
- 18. A unified framework to combine disperate data types in species distribution modelling Data Whole Region Expert range Park, absent Park, present eBird GBIF expert range 2 point processes (49 points) 28 parks
- 19. A unified framework to combine disperate data types in species distribution modelling A Fitted Model mean sd Intercept -0.03 0.02 b.eBird 1.54 0.39 b.GBIF 1.54 0.24 Forest 0.00 0.01 NPP -0.01 0.01 Altitude -0.01 0.01 DistToRange -0.01 0.00
- 20. A unified framework to combine disperate data types in species distribution modelling Predicted Distribution Posterior Mean −0.10 −0.09 −0.08 −0.07 −0.06 −0.05 −0.04 −0.03 −0.02 Posterior Standard Deviation 0.01 0.02 0.03 0.04 0.05 0.06
- 21. A unified framework to combine disperate data types in species distribution modelling Individual Data Types eBird GBIF Parks Expert Range
- 22. A unified framework to combine disperate data types in species distribution modelling Join the bandwagon! Using continuous space - makes life easier In practice, use INLA (but I need to tidy up the code)
- 23. A unified framework to combine disperate data types in species distribution modelling Not the final answer... http://www.gocomics.com/nonsequitur/2014/06/24