Identification of the leaves of Barro Colorado Island
Project summary for Tomas Pickering's STRI intern
Mireya Correa, Stephen Hubbell, Steven Paton, John Pickering, and Joseph Wright
15 February, 2006
This summer Tomas Pickering proposes to build an on-line guide to the
leaves on Barro Colorado Island and collect material to evaluate different
identification technologies.
The summer's goal is for Tom to assemble, scan, and voucher thousands of
leaves from Panama. He will build an IDnature guide on Discover Life as he
does so. Among the people who will help is Osvaldo Calderon, who has spent
the last 20 years identifying a million things that have fallen into
leaf/flower/fruit traps run by Joe Wright's ESP project. Osvaldo can
quickly identify leaves to over 500 species of the island's 700+ species.
Tom will work with Osvaldo and try to put Osvaldo's expertise on-line. In
return, Tom will train Osvaldo on how to use software tools on Discover
Life so that Osvaldo can continue to refine the guide after Tom returns to
Georgia.
Tom will also work with Mireya Correa, Carmen Galdames, and other
botanists, such as the CTFS team working on BCI's 50-hectare plot. Robin
Foster at the Field Museum, for example, has offered us use of all his
images. Steve Paton, the ESP's terrestrial data manager, will supervise
Tom and help him with photography and processing images. Tom will
opportunistically take photographs of fruit, flowers, and bark to add to
Steve's collection of plant images.
Discover Life will provide barcode labels and data management tools for the
samples and images. They will also provide technical support and training.
Ashley MacDonald, who has built an IDnature guide to over 1,200 North
American trees and shrubs will start the process when she goes to Panama
this March to help train people at STRI how to build guides and get images
and other data on-line. Bob Magill and other botanists at Missouri
Botanical Garden will help in integrating an electronic version of the
published Flora of Panama and information from their Tropicos database.
In addition to building an IDnature guide on Discover Life, we would like
the project to compare the ability of various identification technologies
(interactive keys, published dichotonous ones, image analysis software,
genetic barcodes, etc.). Steve Buchmann in Arizona has offered to scan a
limited number of leaves at very high resolution (from 72 dpi up to 5880
dpi) to evaluate how image quality affects the ability of image recognition
software to identify leaves. We plan to send a set of high-quality images
of the undersides of 20 leaves from each of 100 species to Norm Platnick
and Kim Russell at AMNH. When funds become available, they will use these
images to train and test their neural net identification software. Thus,
we hope to test various methods in identifying leaves relative to other
technologies. In short, it's a challenge. Humans versus machines.
Novice users versus experts. High tech versus low tech. Expensive versus
shoe-string equipment. It's a winners-get-a-bottle-of-wine challenge.
And, with luck, there will be a lot of publicity and a methods paper in
Science for us all.
The invitation to participate is open to all serious folks. Eventually we
hope to evaluate Lucid, Delta, and genetic barcoding technologies too.
When Tom processes the leaves that Osvaldo, he, and others collect, he will
preserve a subset for genetic work. He will also collect additional
samples that we can use to test each technology, i. e., when exam time
comes, each team will be challenged with a 100 unknown leaves, including
species that they have not seen before. They will be asked to identify
species and recognize new ones. We plan to get an independent review panel
of judges who will set up the criteria for success and make sure the
competition is fair. We will come up with various metrics to compare each
of our technologies and evaluate the pros and cons of each.
Jeff Lake, a graduate student of Steve Hubbell, is comparing leaf
morphology and has considerable experience scanning leaves. He will help
us develop the best protocol to scan the leaves. Tom will database the
scanned images and photographs and put them on Discover Life servers.
Thus, Jeff and other web users will have access to them.
If all goes well, by the end of the summer, we will have tested an on-line
guide to the leaves of 300 - 500 species on BCI. If the system works,
then experts in Panama can add more species. We will also have a large
number of scanned images, vouchered leaves, and material for DNA work.
These will permit us to compare the effectiveness of different
identification technologies.
|