1 of 8

Using CODAP, Datasets, and AI to Study Seabird Restoration

Caitlin Harrigan - Lyman Moore Middle School, Portland, ME

Jacob Sagrans - Tumblehome, Inc., Boston, MA

NSTA 2024 New Orleans National Conference

November 8, 2024

Link to slides: https://bit.ly/PuffinsNSTA2024

Link to handout: https://bit.ly/PuffinsNSTAhandout

Photos by Derrick Jackson

2 of 8

Puffin burrow cam highlights

(Summer 2023, Seal Island, Maine)

Source of video: https://bit.ly/2023BurrowHighlights

3 of 8

Introduction to the Puffins project

National Science Foundation (NSF)-funded project in Maine middle school science classrooms, in partnership with Tumblehome and Audubon
Atlantic Puffins in Maine is used as a lens into ecology, data science, and artificial intelligence
Significant time spent exploring historical data about puffins in Maine in CODAP (Common Online Data Analysis Platform)
Focus on artificial intelligence—e.g., using Teachable Machine (TM) software to classify images and sounds of puffins and other seabirds

5 minutes for this slide/introductions. CAITLIN DOES THIS (JACOB ALSO INTRODUCES HIMSELF)

Mention that Atlantic Puffins are a charismatic seabird that breeds on islands off the Maine coast (and elsewhere) that were nearly hunted to extinction in Maine 100 years ago, but starting around 50 years the national Audubon society, under the lead of Dr. Steve Kress, began restoring puffins to islands off the coast of Maine–and now they are thriving, but they are facing new threats, like climate change.

Mention that CODAP and Teachable Machine are tools designed for education, appropriate for middle school and up, and are free and run in your browser.

We will be exploring how we use CODAP and Teachable Machine for the Puffins project in this workshop–and we encourage you to think about how you could apply what you learn to phenomena of local relevance to your students.

4 of 8

Exploring historical puffin data in CODAP

Link to historical puffins data in CODAP to explore in pairs/small groups: https://bit.ly/PuffinsReturnCODAP

What factors influence whether puffins reach adulthood and return to their hatching islands to breed?
Put your graphs and answers to the questions in this slideshow: https://bit.ly/PuffinDataSharing

Lots of CODAP help available at: https://codap.concord.org
For more on CODAP, there are 2 CODAP workshops tomorrow (Saturday)!

5 minute introduction, then 15 minutes exploring the data in small groups, followed by 5 minutes looking at graphs (JACOB INTRODUCES THIS)

Mention that this dataset presents information on a sample of 254 puffin chicks (or pufflings) on two islands off the coast of Maine (Seal Island and Matinicus Rock), with hatching years ranging from 2010 to 2018. There are 9 variables, with a mix of categorical, nominal, and numeric data. The purpose of this dataset is for exploring what might contribute to a puffin reaching adulthood and returning to its island of birth. The puffin chicks are banded and then researchers on the islands note if they are seen again in subsequent breeding seasons. Note that sea surface temperature, herring in diet, and butterfish in diet are averages across an entire island in a given breeding season, while weight, wing length (for one wing), and returned as adult are specific to each puffin.

Also there are a lot of help resources available for CODAP (visit the CODAP website at https://codap.concord.org/)

5 of 8

Teachable Machine (TM) and AI – Images

TM general training/testing link: https://bit.ly/TeachableMachineTrain
Create image model with 2 classes (Puffins vs. Other Seabirds)
Train model with 10 puffin images and 10 other seabird images
Then test with 5 different puffin images and 5 different other seabird images
Record if TM accurately classified each image on your handout (confidence above approx. 70%).
How many puffin images did the model accurately classify? How many other seabird images did it accurately classify?

CAITLIN INTRODUCES (15 min total)

Demo how to train/test teachable machine with images (5 min), then have people do it in pairs/small groups (10 min). Say they can use the template to keep track of which images are for training vs. testing. Save some time at the end to talk about why different groups/people got different accuracies. (people may have used different images)

Why do this? Say we were inspired to do this by the puffin burrow cam–it’s impressive technology, but do people really want to watch it 24/7? AI is getting better and better-–what if it could be used to go through the images from the puffin cam and tell when a puffin is in the burrow?

Jacob: Write down on a sheet of paper what accuracies people had

What were things you did that were a success or that you could improve on for training?

6 of 8

Teachable Machine (TM) and AI – Sound

TM can be trained on different classes of sounds, such as “puffins” vs. “background noise”
TM converts sounds to images (spectrograms) to train/test the model
Pre-trained sound model: https://bit.ly/PuffinSoundTM
Try to growl like a puffin (make sure microphone input is toggled on)
Does TM classify you as sounding like a puffin?

Puffin growl

recording (repeated

10 times) & corresponding spectrogram image

JACOB LEADS THIS (POSSIBLY SKIP THIS IF WE DON’T HAVE ENOUGH TIME)

10 minutes for this

TM can do classification of sounds too! Then you will all try to see if you can “growl” like a puffin–does TM think you sound like a puffin.

TM converts sounds to spectrograms, or visual/graphical representations of them, and then uses these to train the model (in a sense it is converting sounds into an image model). Example of spectrogram is seen here. Higher up on the spectrogram is higher tones, and more yellow/red is louder. Let’s play the corresponding sound so you know what a puffin growl sounds like!

The teacher guide for instructions on how to train the model yourself (See day 9, activity 4): https://bit.ly/PuffinsGuideRequest

Now you will all open the pre-trained model and at https://bit.ly/PuffinSoundTM . Try to growl like a puffin (make sure you toggle the microphone input to on if it isn’t on already). Does TM classify you as sounding like a puffin?

7 of 8

Conclusion

A compelling, locally relevant phenomenon (puffins in Maine), paired with accessible tools (CODAP and TM), can engage middle school students in ecology, data science, and AI.
What phenomena might engage your students, and how would you explore using CODAP and TM?

Data on local species in CODAP
Identifying images/sounds of local species
Birds local to you could be a good choice (lots of data/images/sound recordings available online – check out Cornell Lab of Ornithology, eBird, Audubon, etc. for specific bird species)

8 of 8

Thank you!

Caitlin Harrigan - harric@portlandschools.org
Jacob Sagrans - jacob.tumblehome@gmail.com

This project is funded by the National Science Foundation, grant no. DRL-2241777. Any opinions, findings and conclusions or recommendations expressed in these materials are those of the authors and do not necessarily reflect the views of the National Science Foundation.