Forecasting the future - Dr Milan Klöwer is coding change in climate modelling
Dr Milan Klöwer’s enthusiasm for problem solving and climate science has led to his use of AI to develop new models. An Eric and Wendy Schmidt AI in Science Fellow, Milan has created his own software to present a new way of tackling the climate change problem.
It was a curiosity about the world which drew Milan to science, and a high school project which led to him to want to study climate change. “I found it super exciting to think through the various climate feedbacks. Does more CO2 mean more plants, do they then take up all that excess CO2 again?” From this project, he realised some of his questions could be answered by building up his mathematics and physics skills at school. Milan then went on to study meteorology and physical oceanography during his undergraduate degree, which enabled him to start the process of learning how everything in the climate system is connected. This connectivity has influenced Milan’s life, and he is able to be creative, flexible, and free-thinking in his work. Having worked on oceanographic research vessels, and spending a winter in Svalbard studying meteorology in the Arctic, the combination of fieldwork with theoretical and computational work has allowed him to see the bigger picture, and to be able to better understand the climate system as a whole.
Milan thrives when problem solving, and he is particularly excited that he can use his programming and coding knowledge to help with this. He says “Writing code can feel very satisfying as you have created something new that wasn’t there before, and maybe more importantly, can be reused or copied as many times as you like. If this is now something that’s useful for me, for others, for science and hopefully also eventually to society, answering important scientific questions of the world — just by pressing a button — that’s incredibly exciting.”
Incorporating AI into his research wasn’t always easy for Milan. He still does a lot of his work based on the first principles of physics, maths, and computer science, and he acknowledges that it was challenging to let go of some of the ‘old-school’ ways of thinking. However, he recognised the insight that he gained when realising which principles still applied, even when using AI. “Many laws can’t be bent by AI. My tip is therefore not to assume that AI methods are silver bullets. Figure out which information is crucial for a prediction, which information should be withheld, and how does that information flow through your AI method towards the output. Understanding these parallels to conventional approaches is absolutely crucial to use AI successfully. In the end, I’m thinking about AI more as a powerful tool to solve optimisation problems, where most of the work still goes into defining the problem, deciding on predictors, and distilling the more general knowledge from a good model.”
AI has changed Milan’s research trajectory, but he acknowledges that he would still be conducting the same research without AI, it would just be done differently. In his research field, hybrid physics-AI weather and climate modelling is a lot about finding and exploring the balance between what should be done conventionally, and what should be done data-driven using AI. A key turning point in Milan’s work was when he used a software package in 2017 that was written in the Julia programming language. It led him to start writing his own software in Julia; at the time, he didn’t know that this would have a huge impact and determine his career path. “The limits of my language means the limits of my world, as Wittgenstein said. I believe this is also true for programming languages.” Milan has since built SpeedyWeather.jl, a modern intermediate-complexity atmospheric model. With this model, he aims to reinvent atmospheric modelling towards interactivity and extensibility, accelerating climate research.
Milan is looking forward to starting the next step in his career, as a Natural Environment Research Council (NERC) Independent Research Fellow. This will enable him to build his own research group, with a larger focus on supervision of students and grant writing.
“All of these are very exciting new opportunities, but I also want to continue being a scientist, developing atmospheric models by writing code, and solving weather and climate problems through computing” Milan says.
