From surf to satellites: Campbell Watson is bringing AI to Earth science

Long before Campbell Watson began working on climate models and environmental impact reporting for IBM Research, he was an accountant. He was good with numbers, and his parents told him accounting was a good, stable profession, so he joined an accounting firm right after college. He spent his days going to work on the 35th floor of an office building in Melbourne. Watson wore a suit to work every day, where he did tax returns for rich families. By all appearances, he was successfully making his way in the world.  

There was just one problem: He hated being an accountant. He took some solace in early mornings and weekends with his favorite pastime, surfing, to counteract the monotony of spreadsheets and pivot tables. It only took six months before he decided he had to get out of there. Fortunately, going to university in Australia was comparably affordable at the time, so Watson went back to school, starting off in general sciences and ending up in atmospheric science. Today at IBM Research, he works on advanced models of the atmosphere, but back then his interest in the Earth was more about having fun. Now wearing colorful vintage shirts to work, flanked by a virtual background of crashing waves on video calls, Watson was drawn to the branch of science that made his favorite hobby possible. 

“I thought that connection between atmospheres, oceans, and surfing was super cool,” he says. When Watson was 10 years old, a friend taught him to ride the waves. He grew up in Melbourne, near some fantastic beaches, popular for water sports. “I started out on a boogie board, but I was jealous of him being able to stand up, so I went and got a surfboard” he says. In high school, he began surfing regularly with a group of friends. They rode the waves up and down the so-called Surf Coast of Victoria, at spots like the picturesque Bells Beach. New York City’s waves pale in comparison to those Australian ones, and Watson, ever on the search for the perfect wave, once wrote for Vice about how he surfed a perfect artificial wave in inland Texas. His longtime group of surf pals still tries to get together for regular surf excursions to this day.

As a climate and sustainability scientist at IBM Research, the team Watson leads is primarily focuses on geospatial data, which includes a partnership with NASA to model climate change and predict weather. The rest of their time is spent on environmental social governance (ESG) reporting, using large language models (LLMs) to help businesses more efficiently track and report their environmental impact, such as their greenhouse gas emissions throughout the supply chain. 

Watson moved to the United States in 2012 for a postdoctoral fellowship at Yale University where he studied the physics of clouds over Dominica and gravity waves over New Zealand. He has worked at IBM Research for over a decade, after moving from Connecticut to New York in 2014. 

A winding path

In school, Watson had always excelled in science. He liked English, though he tended to get poor grades in the subject. “I just found science easier than humanities, but I was encouraged to go into business.” His mother worked as a secretary and his father was in business recruitment. Family influence aside, though, Watson always felt like his brain was wired to study physics, biology, and other hard sciences. “I wasn’t initially drawn to science as a discipline, it’s just that the information was easier to understand,” he says. “I don’t understand grammar; it’s always been so difficult,” he says. For Watson, his difficulty with language education made sciences seem more intuitive by comparison.

Ironically, he now works with language all the time. But these days, it comes in the form of LLMs. A typical day for him starts with progress report meetings with a few different teams who are working on different scientific modeling problems. These are related to the geospatial models they’re working on, as well as their work on LLMs for ESG reports.  

Off-the-shelf LLMs aren’t very good with things like acronyms and semantic relationships of specific domains in sustainability — things that ESG reports are full of. So, Watson’s team has been using thousands of existing reports to perform continual pre-training on IBM Granite LLMs. They incorporated fine-tuning, instruction-tuning, and RAG to power a chatbot that can perform ESG tasks like crafting report entries and calculating the carbon footprint from supplier transaction data.

One of the challenges they’ve encountered is getting a model to read tables, albeit very different ones from those Watson used to contend with in his accountant days. It sounds simple, but while LLMs are good at parsing a paragraph, most models can’t necessarily tell that one column in a table contains information related to information in another column, or that they’re organized by labels at the top of each column. They’re working on alignment to address that detail, but ultimately, they want to come up with a system that allows interaction between this LLM and the geospatial models. “This could be CO2 emissions from industrial plants, methane emissions from cows, or deforestation from the procurement of palm oil,” Watson says. “This information needs to be represented in these reports, and it can be gathered through geospatial data.” 

For this AI issue, there’s a lot of tricky problems to solve along the way, but he sees merging these processes in some sort of agentic system as the ultimate promise of using models for climate and sustainability. “I think it’s really cool that we’ve used continual pre-training to make the Granite LLM sustainability-aware, but we haven’t adapted it so much that it’s no longer a Granite model,” Watson says.  

Beyond the ESG work, geospatial models occupy much of Watson and his team’s time. This year they reached a milestone with this work, releasing Prithvi WxC in collaboration with NASA. It’s a new general-purpose AI model for weather and climate. “We were tired of building a new model every time we wanted to learn something new fromgeospatial data, so when foundation models came along, we saw them as a great opportunity to address a range of problems,” he says.

Fortunately, a team at NASA was interested in using foundation models for Earth science, weather data, and climate projections. Pulling data from various satellites, which take images at different times, on different scales, and even in different wavelengths of light altogether, has proven to be a challenging multi-modal problem. “It’s a very interesting problem that, if solved, will help a lot of other domains,” Watson says. Prithvi WxC, which is open-source, is a prime example of how his team takes all their research from the basic concept, through to building blocks and then into something that’s actually going to make an impact on the world.  

“We have researchers, developers, scientists, and engineers who are doing this work, but it then needs to be translated to be something useful for our partners, collaborators, and the community,” he says. This translation step is key, as it differentiates the fundamental research being done at IBM Research from similar work at your typical academic institution. Part of Watson’s day-to-day work is dedicated to mapping that process, making sure that the research is moving in a direction that ensures it’ll be useful to businesses and other organizations, and that those who could benefit from his team’s work either understand what the team is doing, or provide input to move it in that direction. 

Learning to lead

As Watson has transitioned into a lab leader these past few years, he’s in a position to define strategy. His role is to question whether a project is scalable, useful, and impactful — and to do something about it if he thinks it isn’t. After all, you don’t just grab your surfboard and head to the beach without checking the weather report, finding out which beach has the best conditions that day, and consulting tide charts. Nor do you drop into a wave without a plan. Growing into this role, Watson was surprised to find he enjoys learning leadership skills. “ It’s been fun to discover my own leadership style,” he says. “Although I can think of times in the past when I found myself leading things, it’s felt somewhat accidental.” These data points reassured him that it was possible. 

Some happened outside of work, in situations where Watson was genuinely surprised at his ability to pull people into a project just for fun. One of these projects was a live coding performance where he wrote code in Ruby and Python, which was translated into sound. The result was equal parts art and science: He installed a weather station atop the Ace Hotel in New York City, and he performed his musical piece while a friend did the live coding to produce visual accompaniments. To pull it all off, they needed other folks to pitch in to set up sound equipment, solder the microelectronics in the weather station’s air quality monitoring system, and make sure the weather station’s internet connection was consistent.

“It was really interesting to see how willing people were to help,” he says. “What are you getting out of this? Why are you saying yes?” he would wonder. But at the same time, Watson acknowledges that he’s eager to lend a hand when asked.  

In surfing, a perfect day means keeping your head above water and enjoying each wave that the ocean gifts you. In his work at IBM Research, Watson has found joy and meaning in the journey, the daily effort that it takes to build on yesterday’s wins and bring his team along for the ride — and of course, the satisfaction of making things work.