Taylor Shropshire started out trying to forecast ideal ocean fishing spots for one of his clients. He ended up on a mission to protect endangered North Atlantic right whales.

Shropshire – who is Head of Marine Resiliency at Fathom Science, a startup that is building a digital twin of the ocean – reasoned that the same types of data he was using to locate fish could just as reliably be repurposed to help mariners avoid accidental collisions that could kill the remaining right whale population.

“Anytime you have good observations of any event that happened in the ocean, we can create statistical models about where they are likely to happen again in the future,” Shropshire says. “So when I saw that there was this issue with the right whale and I saw there was great historical data on it, we started looking to see how difficult it would be to come up with a forecast of risk.”

In October 2024, NOAA Fisheries estimated that there were 372 (+11/-12) individual North Atlantic right whales alive during 2023. The annual population abundance and death estimates displayed in this infographic were calculated using the model described in Linden 2024. The documented birth data displayed in this infographic reflect the number of mother-calf pairs that were visually sighted in the Southeast US through the most recent calving season. Credit: NOAA Fisheries

Fathom Science combined its rich ocean data with historical whale sighting data to create its WhaleCast – a heat map visual that looks something like an hourly weather forecast, making it more dynamic than static maps of government-regulated “seasonal management areas.”

While Shropshire is a data scientist in addition to an oceanographer, he lacked a pure knowledge of statistics and machine learning, which is where the SAS Data for Good program came in.
 

SEA-ING IS BELIEVING

Lincoln Groves, a SAS employee volunteering on the project, was tasked with validating WhaleCast’s predictive ability by building a number of machine learning models.

Not long after undertaking the project, Groves and family took a trip to the North Carolina Aquarium at Fort Fisher on the North Carolina coast. And what did he encounter but an exhibit on North Atlantic right whales. It was serendipity.

Here he was, smack-dab in the path of the right whales, who move seasonally between roughly Georgia and Maine, at the edge of a part of the ocean where – theoretically – there might be whales that very minute. And where, once upon a time, there might have been many more of them.

“It was really kind of cool to see that this is literally part of the migration pattern – this is what we’re trying to predict, and here’s why it matters,” Groves says.

A visualization illustrating the risk right whales face every day. It shows the migratory path of a 1-year-old right whale satellite-tagged off the Virginia/North Carolina coast in March 2021. Credit: NOAA Fisheries.

Right whales have been on the endangered species list since 1970. While the whaling that decimated their populations in the 1890s is no longer an issue, boat strikes and equipment entanglements still pose a grave threat – and have picked up since 2017. Today, there are approximately 370 individuals remaining.

SYNTHETIC DATA, REAL PROGRESS

Groves was ready to try out a series of models to see if he could validate the WhaleCast, but there was something he needed first – a whole lot more data.

To avoid what data scientists call “overfitting” in machine learning, you would ideally use much larger data sets than the 40,000 points of data Fathom shared. So Groves used SAS® Data Maker to create synthetic data – basically, manufactured data with the characteristics of the actual data – to expand his set to 500,000 data points and separate them into buckets for training, validation and testing.

“It was pretty interesting to see him create multiple models very quickly,” Shropshire said. “He was kind of able to go from the very simplest model to really complex kind of neural network, machine learning-type models and show the benefits and the limitations of each.”

Groves then decided to use SAS® Viya® Workbench, a standalone programming environment, to creatively solve a second problem: the need to calculate the probability of whales’ distance from shore. In Workbench, he was able to pull in a Python package, quickly do the calculations and kick the results back over into his workflow.

Groves’ combined work resulted in Fathom Science now having validation from a statistical and machine learning perspective that their approach to predicting where right whales might be was a good one.

“If you get the same answer two different ways,” Shropshire says, “then you feel more confident about it,” Shropshire says.

‘CAN WE EMPOWER PEOPLE?’

Governments already establish seasonal management areas where ships of a certain size have to keep their speed down along the Atlantic coast, but that paints with a pretty broad brush for a shipping industry often forced to choose between commerce and conservation.

And other voluntary technologies for avoiding right whales, like thermal cameras, can be prohibitively expensive for ship owners.

The WhaleCast, though, is designed to be integrated with existing on-board touch screens that put resources like weather maps at mariners’ fingertips, so they can better understand where risks are higher or lower.

“A lot of mariners have felt kind of controlled top-down by the government imposing all these restrictions,” Shropshire says. “What we’re trying to do is ask, can we empower people? Right now, mariners have very little available to them to make these kinds of dynamic decisions, so if you provide information, you can increase people’s sense of efficacy.”

It’s a project that neither Groves nor the team at Fathom envisioned partnering on until the opportunity suddenly presented itself to tackle a practical problem by bringing together the unlikely combination of ocean data, AI and a species in need of a lifeline.

Once viable models are established from digital twins and machine learning, you can expand how and where to use them, meaning the sky – or, in this case, the ocean – is the limit.

“I think that there’s a lot of hype with machine learning models, and there’s a lot of hype when we talk about AI,” Groves says. “But when you talk about something like right whales, people understand what we’re trying to do here.”