14
Crowd-sourcing ecology: Predicting plant attractiveness to pollinators from internet image searches Christie Bahlai and Douglas Landis Michigan State University

Crowd sourcing ecology: using the internet to develop hypotheses about pollinator preferences

Embed Size (px)

Citation preview

Crowd-sourcing ecology: Predicting plant attractiveness to pollinators from

internet image searches

Christie Bahlai and Douglas LandisMichigan State University

Restoring landscapes to support pollinatorsUse plant communities that are:

Attractive DiverseLocally adaptedNative

Common garden experimentsSoil typesLocal climate

Is there a better way?Goal: identify candidate plants for locally

customized habitat restoration

Can we use existing data to help narrow our search?

Hypothesis People like to take pictures of plants in bloom People like to post pictures on the internet Some of the pictures will capture insect visitations

Plants that are highly attractive to pollinators will be photographed being visited by pollinators more frequently

Ellen Booraem, http://ellenbooraem.blogspot.com/

ApproachDetermine search terms and engines to useSearch for images of plants with

experimentally known pollinator visitation rates

See if relative visitation rate observed in searches ‘predicts’ relative attractiveness of flower

Search terms and enginesPlant species Search term

Search engine Latin name Common name

"Beneficial Insect"

"Insect" "Bee"

"Honey Bee"

Bing          Vicia faba Fava Bean 0 (7) 0 6 1 (17)Fagopyrum esculentum Buckwheat 0 (7) 5 8 3Coriandrum sativum Coriander 0 (19) 0 3 0 (27)Lobularia maritima Sweet Alyssum 1 1 3 4Anethum graveolens Dill 0 0 1 0 (21)Total 1 6 21 8

Google

Vicia faba Fava Bean 2 3 11 8Fagopyrum esculentum Buckwheat 1 3 7 4Coriandrum sativum Coriander 1 4 2 2Lobularia maritima Sweet Alyssum 1 1 2 2Anethum graveolens Dill 1 3 0 0Total 6 14 22 16

Testing the associationExisting surveys:

Tuell et al 2008 (Apis and non-Apis bees)

Fieldler 2006 (Syrphid flies)

Search Google images for “[plant species] bee” Record number of images with each taxa in them

Photos: John Severns, Wojciech Ochwat, Kevin Hall

Results

ResultsRelationships between

visitation rates in field for non-Apis bees, syrphids

No relationship for Apis bees

Model

Slope (field/

images) Pseudo-R2

Apis bees - -

Non-Apis

bees*

0.10±0.04 0.668

Syrphids 0.08±0.72 0.003

DiscussionNeat! It worked (for non-Apis bees)!Why were no relationships observed for Apis

bees?

Photos: John Severns, Wojciech Ochwat, Kevin Hall

Applications“Passive crowdsourcing”

More informationSee complete data+ analysis for this

experiment on GitHub (cbahlai/Bee_images)

Thank you!Data collection: Julia PerronePhotos: Carolyn Malmstrom, Ashley

Bennett, Rufus Isaacs