SkyTruth’s machine learning engineer, Jona Raphael, grew up in Shepherdstown, West Virginia; SkyTruth’s home base. But it was his travels around the world that convinced him to knock on our door.
Jona Raphael is privileged. It’s a phrase he repeats often when discussing why he does what he does. “As an American, I am in the wealthiest one percent of the globe. But that privilege comes with responsibility,” he says. For Jona, this responsibility translates into sustainable design, whether it’s developing a financial system for the poorest of the world to pay for their daily (renewable) electricity, or giving shrimp farmers a competitive edge over trawlers that destroy the seafloor. At SkyTruth, it translates into developing Cerulean, a machine learning model that automatically processes thousands of satellite images a day to detect oil pollution in the ocean, protecting marine life and coastal communities, and highlighting the dangers of relying on fossil fuels.
“I’ve always had an interest in sustainability,” Jona says. “Probably the root cause was traveling with my parents as a child.” Jona grew up in the small town of Shepherdstown, West Virginia, home of SkyTruth’s headquarters. In his childhood travels with his family, he observed how other people lived in countries around the world – including many, if not most, who were less privileged than he was. “Seeing the diversity of human experiences created a deep-seated empathy in me,” he says.
But his route to sustainable design and environmental protection was a bit circuitous. As a high school student, Jona loved math and science but was also interested in anthropology and religious studies. He couldn’t make up his mind about what to study. So he applied to two very different colleges: the old and prestigious Brown University and the new, unaccredited Olin College of Engineering. He spent his freshman year at Brown, pursuing his interests in the humanities. Then, he indulged his math and science side at Olin College his sophomore year. There he found a very small and technically minded community that challenged his thinking. The pace was fast— assignments were due constantly, not just at the end of the semester. The school embraced the idea that failure was part of learning. From the start, students were required to give presentations of their work. “We got feedback and we got graded” on those presentations, Jona says. “Everyone started bad and got better, so by senior year we were all really good. We could pitch $50,000 or $100,000 projects to potential funders.” Moreover, at this new college, students helped develop the curriculum – another plus in Jona’s mind.
Jona also met his partner Becky Belisle that first year at Olin. There was no going back to Brown.
He majored in mechanical engineering – one of the few options available at the small school – and spent a summer volunteering for an agricultural nonprofit in rural India where he helped develop a lantern that ran on ethanol and water; resources available to local farmers. After graduation and a few design jobs, Jona returned to school and completed his master’s degree in mechanical engineering at Boston University in 2012.
“My brother was the coder,” he says, explaining why he didn’t initially go into programming. “I avoided coding for fear of being in his shadow or falling in his footsteps. Turns out that was a bad decision.”
Or not. Jona began programming out of necessity in one of his first jobs after graduate school, at a San Francisco startup called Lumeter. Lumeter connected people living in poverty around the world with renewable lighting sources by creating a manageable payment plan that allowed them to pay for lighting as they needed it. “We wrote an entire banking system that supported micro payments from tens of thousands of the most impoverished people around the world – people in very remote areas,” Jona explains. “How do you collect 10 cents each from all these people? Most of that solution actually exists in code.” And in electronic circuit boards. Over time, Jona took more and more independent study courses in coding and machine learning – at least a dozen so far, he estimates – teaching himself how to train computers.
Along the way, he and Becky got married and spent four months traveling the world for their honeymoon: scuba diving in Indonesia, trekking in Nepal, bicycling across Rwanda, rock climbing in Vietnam, and more. “Becky loves the outdoors even more than I do,” Jona says. “And so rock climbing is kind of a compromise. Becky could just hike and hike and hike but I like more of a game.” Rock climbing is sufficiently “game-like” for Jona.
On their honeymoon in Indonesia, Jona and Becky gained an appreciation for the joys and challenges of remote island life. Jona took this photo in Raja Ampat.
His route to SkyTruth began while working with the startup Osmo addressing aquaculture issues. The Osmo crew had turned to Global Fishing Watch (GFW) for data to help guide their business decisions and attract funders and users. SkyTruth is one of GFW’s founders. When Becky got her dream job as a physics professor at Wellesley College, they moved to Boston and Jona was free to pursue his passions for machine learning and sustainability as a contractor. Searching for potential clients, he returned to the GFW website and noticed that SkyTruth was a partner. “That rings a bell,” he thought, remembering his Shepherdstown roots. He reached out to SkyTruth’s Chief Operating Officer, Jenny Allen, about how he could help. Soon, he was learning how to analyze radar satellite images to detect oil slicks at sea and developing a model that automatically recognizes and flags these events. (You can learn more about Jona’s process for training computers to analyze thousands of images a day and detect oil slicks in his blog post here.)
Cerulean has been collecting data about apparent oil spills for over a year now. Among the thousands of spills Cerulean has flagged, it has positively identified over a hundred specific vessels in close proximity at the time of detection, pointing to the likely polluters. With Cerulean, SkyTruth hopes to set the global standard for oil slick detection and establish a ground truth for activists and policymakers to use when navigating hard decisions related to oceanic pollution and climate change. In the near future it will be expanding to detect spills and methane flaring from offshore oil platforms. Ultimately, Cerulean will allow SkyTruth’s conservation and watchdog partners to hold polluters accountable, change shipping practices, and stop offshore oil drilling to protect the world’s climate.
SkyTruthers Ry Covington, Jona Raphael, and John Amos presented the challenge of oil slick detection to participants at the AWS re:Invent Hackathon for Good in Las Vegas in December, 2019. Many teams at the hackathon generated creative uses and presentations to help advance Cerulean. Photo credit: unknown.
“I really am an evangelist about what machine learning can do,” Jona says. “There is no doubt in my mind that for-profit companies are pouring resources into Artificial Intelligence systems and reaping the benefits, whether through number of clicks, number of sales, or gallons of oil guzzled. I saw the nonprofit community flailing, being left behind, still trying to solve problems with brick and mortar solutions, when I think that so much can be solved algorithmically through machine learning. I made it my objective to find an organization where I could bring that vision.”
In his passion for machine learning, Jona sounds a lot like SkyTruth’s President John Amos, who founded SkyTruth back in 2001 to help conservation nonprofits protect the planet with the same satellite imagery that extractive industries were using to destroy it. Twenty years later, thanks to Jona’s expertise, SkyTruth is now marrying satellite imagery with the latest computing technology to exponentially expand our impact.
And for all of us at SkyTruth, expanding our impact is a tremendous privilege and responsibility.