Name that Bumble Bee: Maybe machines can get pretty good at identifying images of bees
This summer has been a pretty exciting year for bumble bees here on the farm. Historically, common eastern bumble bees have been the exclusive bombus of the realm. But this season, there were two new types spotted among the wildflowers and gardens – both distinctly different from impatiens in their coloration and (unless my ears and memory deceive me) their flight buzzes. The first was identified by readers of the Bee Report as fervidus, the yellow bumble bee. The second remained a mystery up until this week, when I did a little experiment.
To identify this second bumble bee, I turned to the expert bee people of Twitter. But I also tried out the BeeMachine – a new deep learning web app for identifying bumble bee species from photos.
Using the BeeMachine is super simple. Take a photo of the bumble bee. Crop the photo as close to the bee as possible, removing as much of the non-bee parts as possible. Upload your photo to the BeeMachine website. Get your results. Instantly. These results are the top three mostly likely species for the bumble bee in question, each with a probability of accuracy. For example, when I submitted two separate photos of my second mystery bumble bee, the results looked like this:
And these results are, quite beautifully, the numerical equivalent of the feedback I got from human experts on the same photos:
• “Probably auricomus. Could be pensylvanicus.”
• “Bombus auricomus, I think! (Watch me embarrass myself with eastern bumble bees...)”
• “I think B. pensylvanicus”
• “...color pattern on this girl is typical auricomus, so I'm pretty confident calling it that”
• “Looks like auricomus to me, but photos are always tricky”
So, bee number two is most likely a black and gold bumble bee. Humans and machine agree on this ID.
I also ran a photo of my first mystery bumble bee through the BeeMachine as another test. And, again, the results were a numerical reflection of what human experts had said: most likely fervidus. Very cool.
However, this experiment does raise an important question: If humans can provide identifications, why do we need a tool like this? What does it gain us?
“Most of the general public, even those enthusiastic about bumble bees, have a hard time identifying them,” says Brian Spiesman, creator of the BeeMachine and a research assistant professor in the Department of Entomology at Kansas State University. This difficulty with identification was recently reflected in a study from York University showing that citizen scientists identified bumble bees correctly only 50% of the time.
“Experts are generous with their time but can get bogged down identifying bees,” Spiesman continued. “Having an algorithm that could lighten their load could allow them to focus more on their own research. It would also benefit citizen science programs that rely on experts to verify user IDs. There are thousands of unverified images at Bumble Bee Watch that are waiting on volunteer experts. Until then, these data can’t be used to study important conservation issues and monitor pollinator populations.”
Of course, machine learning and automatic identification is only helpful for breaking a bottleneck like this if the output is reliable and accurate. Which means the images that the neural network is initially trained on must be accurately identified to begin with. Otherwise, garbage in, garbage out. In this case, the BeeMachine has been developed with images and data from the Xerces Society and Bumble Bee Watch, iNaturalist, BugGuide, and Spiesman's own photo collection.
Spiesman is planning to expand this type of automated identification in other ways in the future, such as deploying automated monitoring systems that record observations of bees visiting flowers. “By pairing cameras with tiny relatively cheap computers (e.g., a Raspberry Pi) loaded with the detection and classification algorithms, we can generate a lot of good data quickly,” he says. “In 2012, I spent hundreds of hours watching video that I collected in the field of pollinators visiting flowers. New automated monitoring technology would do most of this work for me.”
Spiesman also hopes to have a mobile app available by the start of bee season next spring. The algorithm would live on a user’s phone so that it can work quickly and won’t need to be connected to the internet or a cellular network. “I imagine people might want to use this in remote places that don’t have good cell service,” Spiesman says. “The first iteration of the app will be somewhat simple but with more functionality than the current web app. More functionality will be built in over time.”
Are you enjoying the Bee Report newsletter? Maybe your friends and colleagues would too!
The podcast is on a short hiatus! In the meantime, why not catch up on some previous episodes? Available on Apple, Stitcher, other major platforms and at thebeereport.buzzsprout.com.
Do you have tips, comments, questions or ideas for collaboration? Please send them to tbr@bymattkelly.com.
Conservation
Endangered rusty patched bumble bees spotted at Illinois forest preserve
(Quad City Times) A park ranger at Illiniwek Forest Preserve spotted an endangered rusty patched bumble bee in the preserve. The sighting was confirmed by an assistant professor of biology at Black Hawk College. “This is the best possible indicator that the prairie restorations the Forest Preserve District has been working on are paying off. Finding this new sighting is really important in our efforts of staving off the extinction of this species.”
Economics
Sharing wild bees helps more farmers
(EurekAlert, University of Minnesota) Many farmers are used to sharing big equipment – like tractors and other costly machinery – with neighboring farms. Sharing cuts costs, lowers the farmer's debt load, and increases community wellbeing. But big machinery might not be the only opportunity for farmers to reap the benefits of cost-sharing with their neighbors. New research suggests that the concept could also be applied to a more lively kind of agricultural resource: wild bees. "What we're proposing is that those farmers providing bee habitat could be rewarded for doing so, to the benefit of all."
Science
Ecologists report climate change affecting bee, plant life cycles
(EurekAlert, Utah State University) “We find bee emergence timing is advancing with snowmelt timing, but bee phenology - timing of emergence, peak abundance and senescence - is less sensitive than flower phenology. Given global concerns about pollinator declines, the research provides important insight into the potential for reduced synchrony between flowers and their pollinators under climate change.”
A honey bee’s tongue is more Swiss Army knife than ladle
(New York Times) For a century, scientists have known how honey bees drink nectar: They lap it up. Now scientists have discovered bees can also suck nectar, which is more efficient when the sugar content is lower and the nectar is less viscous. And not only do honey bees have this unexpected ability, but they can go back and forth from one drinking mode to another.
(The Conversation) “As behavioral ecologists who have studied social interactions in honey bees, we see parallels between life in the hive and efforts to manage COVID-19 in densely populated settings. Although honey bees live in conditions that aren’t conducive to social distancing, they have developed unique ways to deal with disease by collectively working to keep the colony healthy.”
A 429-million-year-old trilobite had eyes like those of modern bees
(Scientific American) Despite being 429 million years old, this trilobite has a modern-looking eye that resembles those of today’s bees and dragonflies. “For a long time, it was thought that just bones, teeth and other hard objects could be preserved in the fossil record. To be able to distinguish cellular structures, especially in the eyes, is very, very rare and exceptional.”
Society/Culture
118 student memberships funded
(Twitter, Entomologists of Color @EntoPOC) “As of 1 August, your donations have funded 118 student members to ecological and arachnological societies from: Brazil, Canada, Chilke, Colombia, Ecuador, El Salvador, Guyana, India, Indonesia, Kenya, México, Nepal, Nigeria, Norway, Pakistan, Peru, Puerto Rico, Rwanda, South Africa, Spain, Sudan, UK, & USA”
Youth entrepreneur Mikaila Ulmer on learning to “Bee Fearless”
(Marketplace) At 15 years old, Mikaila Ulmer is a student, bee ambassador, social entrepreneur and author. The origins of her company, Me & the Bees Lemonade, which donates a portion of sales to bee conservation, dates back to a lemonade stand Mikaila started in kindergarten for a children’s business competition in Austin, Texas. After 10 years in business and an appearance on the television show “Shark Tank,” Mikaila’s company is a national brand. Her flaxseed lemonade is sold in 1,500 stores in more than 40 states.
One More Thing…
I want to take the bee-equivalent of this photo. Wow. “Young Cooper’s Hawk on the hunt,” from Kitundu @birdturntable via Twitter.