Just back from the Arctic Circle, Google DeepMind Scholar Anna Vaughan (Darwin 2019) is feeling ebullient—even though things didn’t quite go according to plan.

"I'm really interested in the Arctic! I've spent time in Greenland, Svalbard, and Antarctica during my PhD and I love it. I was there to do a course in Svalbard on Arctic renewable energy but ended up working instead, planning for my return in March and May to run my own fieldwork campaign by putting out sensors according to a map created by a machine learning algorithm. We aim to develop a new weather forecasting system for the Arctic because it's a region where traditional forecasting systems really struggle."

But Anna’s passion for weather forecasting comprises much more than the Arctic. 

She’s in the final year of her PhD researching applications of machine learning to climate and weather—made possible by Google DeepMind, a world leader in artificial intelligence research and its application to real-world problems. The Google DeepMind Scholarship fully funds female and/or Black postgraduate students studying AI-related fields—all currently underrepresented demographics—at the University of Cambridge, through the Cambridge Trust. It’s part of the company’s effort to build a stronger and more inclusive AI community and bring a wider range of experiences to the fields of AI and computer science. 

Prior to her PhD, Anna completed her undergraduate studies in pure mathematics and theoretical physics, followed by an MSc in meteorology from the University of Melbourne. Her MRes project in Cambridge’s Centre for Doctoral Training developed a new methodology for downscaling climate model output using Bayesian learning.   

Where did her keen interest in climate, weather and meteorology stem from? 

"When I was four, I got into tornadoes! I had a meteorology book, and I loved the tornadoes. Then I had nightmares about tornadoes for many years because I thought they were going to come to Tasmania, where I’m from—which is unlikely, although one did arrive in 2012. I really got into forecasting when I was eight thanks to Tropical Cyclone Ingrid, which was headed towards Australia. I remember hearing about it on the radio and I was amazed that they could predict where it was going to go. It went right along the top of Australia, and I started downloading forecasts and making my own little predictions. And that was the start of it." 

This year, several of the projects Anna has worked on throughout her PhD have come to fruition.   

One notable example is Aardvark Weather, the first end-to-end data-driven weather forecasting system. This model can be optimised for any region, opening the door to cheap, high-performing weather forecasting systems worldwide including in low-resource areas most vulnerable to the impacts of climate change. Anna takes up the story: 

"I’m very interested in how AI in weather forecasting can translate to improved systems for developing countries and under-resourced areas. For one thing, agencies can develop and run AI systems locally, whereas for a traditional system, you need a supercomputer. These are incredibly computationally expensive and involve millions of lines of code plus complex different modules that slot in. You need teams of scientists to develop them and to change stuff.

"But with AI you can make a simple, modular, lightweight framework that is orders of magnitude faster in generating a forecast. And the way that an AI weather model is set up, it's much easier to deal with aspects like noisy data and diverse inputs. While there is an expensive training phase, there are grants available. After that, anyone can run these models and create bespoke models as well. So rather than trying to run something optimised for the mid-latitudes over Europe and Africa, where a lot of our expertise is concentrated, we can develop something specifically for West Africa, or the Arctic, or other regions where you've got quite distinct meteorology."

According to Anna, the Arctic can confound forecasting due to a lack of observational data—fewer people means fewer sensors—as well as the complexity of processes such as sea-ice interactions, which are often not well understood.  

"Using AI in this case is an exciting opportunity, and there are so many applications for AI models in this area. My work recently has looked at tropical cyclones. There was a paper from a while back talking about how tropical cyclone track forecasts are likely as good as they can get.  And yet, we're outperforming that by 50 per cent with this new AI model! Extremes forecasting like this is what I want to go into deeper in my postdoc. In terms of developing bespoke systems for different extreme events, there is a huge amount of scope, and I’d like to see this applied downstream.

"It’s extraordinary what AI is enabling in this critically important field. These developments represent enormous impact across society, the environment, and the economy, particularly for the planning that this kind of forecasting allows—for example if you can predict the track of a cyclone:

"The timescales I'm interested in are 1 to 15 days. We had one forecast where we got it correct five days earlier than the official forecast. That's five extra days for people to board up their houses, sandbag, move to evacuation centres—particularly for lower-income areas where you may not have great transport links in and out, those extra days are absolutely crucial."
 
To conclude her thesis Anna is currently developing a hurricane-specific data-driven forecasting model and working on several other projects in conservation, seasonal forecasting, poverty mapping and applications of Large Language Models to weather and climate. She plans to continue working on these projects in 2025 as a postdoctoral researcher at Cambridge.  

In Anna’s view, why is Cambridge the place to work on solving these problems? What makes the University special? 

"It’s the people here: my collaborators are insanely talented! There's just no way I would have done any of this work without them—and the brilliant supervisors. There are more people in ‘niche’ areas, too.  I come from a very small university in Tasmania and our departments there were tiny. So to be able to work in more specialist areas the extensive Cambridge community is ideal. Just world-class.  

"I am grateful to Google DeepMind for their support over the past four years. This opportunity to study at Cambridge has opened doors I wasn’t even aware existed. Coming from regional Australia there was no way I would have been able to afford to undertake this course without this scholarship. It has been an immense privilege to study here, and it is inspiring to reflect on the diverse stories and research impacts across this field that have stemmed from the generosity of Google DeepMind’s funding programme."

Thanks to the power of philanthropic giving, rare and remarkable minds like Anna’s have free rein to explore, collaborate, research, and devise unprecedented solutions in the unique, cutting-edge atmosphere that Cambridge provides. From youthful but inspired beginnings, Anna is truly fulfilling her intellectual promise—for the good of the planet and for all our tomorrows.