1 / 21

Ceratina on Dianthus flower

Graphical Models and Pollination. Ayesha Ali University of Guelph. With: Tom Woodcock, Liam Callaghan, Catherine Crea. TIES 2010 June 23, 1010. Ceratina on Dianthus flower. Outline. Motivation: Pollination Ecology Qualitative Pollination Webs

wes
Download Presentation

Ceratina on Dianthus flower

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Graphical Models and Pollination • Ayesha Ali • University of Guelph With: Tom Woodcock, Liam Callaghan, Catherine Crea. TIES 2010 June 23, 1010 Ceratina on Dianthus flower

  2. Outline • Motivation: Pollination Ecology • Qualitative Pollination Webs - Feature Extraction • Quantitative Pollination Webs - Driving Mechanisms • Hierarchical graphical models

  3. Motivation: Mutualistic relationship • Plants need to be pollinated by birds and insects for reproduction • Offer rewards for being visited, (e.g. pollen, nectar, oil) Halictidae on Queen Anne’s Lace

  4. Motivation: Species decline • Recent years has seen a decline in some insect species (e.g. bees) • Forest fragmentation has led to a decline in some plant species Andrena – native wild bee

  5. Motivation: Species decline • Extinction of given plant may adversely affect survival of given insect, and vice versa (e.g. Mauna Kea silversword ) • Need to maintain species abundance / diversity in ecosystem • Ans: Pollination webs? Orthonevra drinking nectar on HopTree

  6. Pollination Webs: bi-partite graph • Nodes are plant and insect species • Edges from insects to plants represent plant-insect interaction • Often called “interaction” or “visitation” web • Only small fraction of interactions observed • Similar to food webs, except role of pollinator and pollinated never change

  7. Pollination Webs: bi-partite graph Pollinators (Insects) Pollinated (Plants)

  8. Pollination ecologist approach • Use adjacency matrix I (N x M) I AF = 1 if animal A visited flower F 0 otherwise • Given a pollination web, what are the important features that characterize the plant-pollinator interactions?

  9. Pollination Webs Pollinators (Insects) Pollinated (Plants)

  10. Ecosystem Interventions • Can we infer consequence of eco-system disturbances (eg. removal of a player due to forest fragmentation)? • Which plants or animals are vulnerable to presence of non-natives? • Problem: • Quantification of connection strength, and • Understanding mechanism behind interactions

  11. Quantified Pollination Webs • Let Xij = frequency of ij-interactions observed • Conditional on the total number of counts, X ~ Multinomial(p) • Proportions are correlated within insect species • Observed interactions are actually a mixture of pollination visits, and non-pollination visits

  12. Quantified Pollination Webs • We can use graphical models to represent the data generating mechanism • Two main issues: How to incorporate • Visit type • Driving force behind interactions? • Use hierarchical graphical model, with probability that an insect-plant pair interact depending on other variables

  13. Hierarchical Pollination Model I • Insects visit one of M floral species, with probability based on the unobserved visit type • Use a variational EM-algorithm to get a generative model of the process, by incorporating the unobserved visit types • Similar idea in AI user rating profile models: • Users rate each of M items, based on some unobserved attitude toward each item

  14. Hierarchical Pollination Model I α p • For each specie: • X | z,p ~ Multin(pz) • Z | θ ~ Bern(θ) • θ ~ Beta() θ X Z M na • Z is an unobserved random variable that is 1 if pollination visit, 0 otherwise • pafz = Pr(insect a visits plant f | visit type z)

  15. Hierarchical Pollination Model I • Free energy maximization (Neal and Hinton) • E-step: compute • M-step: maximize free energy wrt variational and model parameters (fixed-point iteration or Newton-Raphson)

  16. Hierarchical Pollination Model II • Borrow from econometrics choice models: • Consumers assign a utility to each of M items • Conditional on the total number of counts, X ~ Multinomial(p)

  17. Hierarchical Pollination Model II δ β • For each specie a: • X | p ~ Multin(p) • exp(ηjg)| δa ~ • Gamma(δa-1λfa, δa-1) • p ~ Dirichlet(δa-1λa) η p X M w na • p follows a Dirichlet-multinomial regression: • Space, time, phenotypic and/or phylogenetic traits of pollinators or flowers or both

  18. Hierarchical Pollination Model II • Fitting presents no computational issues – Newton-Raphson can converge quickly • Can use existing software to fit model (LIMDEP, Stata, etc.: negative binomial with fixed effects for panel count data) • Vasquez et al. (2009) present a non-stochastic version of this framework

  19. Conclusions • Pollination webs can help to understand insect-floral interactions • Hierarchical models provide a framework for incorporating covariates into the generative model • Provide insights into where conservation efforts should be placed

  20. Future Works • Learn linkage rules: mine bootstrapped samples of data • Overdispersion due to “real” zero-interactions • Modify error distribution for utilities in order to study competition between insects

  21. THANKS! • CANPOLIN • Tom Woodcock • Elizabeth Elle • Peter Kevan Syrphidae Pt Pelee

More Related