Climate modelling between bits and bytes

Hi there. I’m Milan. A climate scientist.

I’m a Schmidt AI in Science Fellow at the University of Oxford working on combining climate modelling and machine learning. During my PostDoc at the Massachusetts Institute of Technology I started developing SpeedyWeather.jl, a modern intermediate-complexity atmospheric model written in Julia. With this model I’m on a mission to reinvent atmospheric modelling towards interactivity and extensibility accelerating climate research. Based on this user and developer-friendly model, I am using machine learning to correct the simulated climate towards data. My Schmidt AI in Science Fellowship aims to do that for precipitation in an online-learning fashion, integrating a machine-learned precipitation correction on every time step.

I am also an incoming NERC Independent Research Fellow at Oxford Physics/AOPP to be started in summer 2025. This means I can supervise Master and PhD (DPhil in Oxford slang) students now, particularly in the PhD programmes Intelligent Earth CDT and the Environmental Research DTP. If you are interested in pursuing a PhD and would love to work on and learn about atmospheric modelling in combination with machine learning please reach out. Especially if you are not keen to drown in dusty Fortran code — not my passion either. I am always fire to bring brandnew concepts of software engineering and computer science into climate research. That’s why I am an active Julia developer, see my GitHub profile. My research interests cover large various aspects between climate physics, modelling, computer science and software engineering, see below.

I hold a PhD from Oxford in climate computing with Tim Palmer after a very comprehensive education in climate physics from different universities in Germany, France, and Norway. During that time, I crossed the Tropical Atlantic on an oceanographic research vessel and spent a winter in Svalbard studying meteorology in the Arctic with weather stations. The combination of fieldwork with the theoretical and computational work of my research has allowed me to see the bigger picture and to better understand the climate system as a whole.

  • Climate modelling: Combining machine learning and dynamical models, atmosphere and ocean, grid-point and spectral, dynamical core development, stochastic parameterizations, turbulence closures.
  • Computing: High-performance, low-precision, parallel, CPU vs GPU, number formats, stochastic rounding, efficiency.
  • Data compression: Real information in climate data, lossy and lossless, information theory, data formats.
  • Predictability of weather and climate: Chaos, uncertainty, ensemble prediction, error growth, weather forecasting.
  • Software engineering: Open source, multiple dispatch and code composability, automatic differentiation, and the Julia programming language.
  • Data visualisation and science communication: Intuitive diagrams, user experience, better posters, better talks, accessibility and alwasy very pretty plots.
  • Aviation and decarbonisation: Global warming impact, non-CO2 effects, emission scenarios, carbon footprints.

As a climate scientist, I feel responsible to voice urgency and hope on the climate catastrophe, both through media and in teaching. While my main research develops computationally efficient climate models, I see great potential in interdisciplinary work to tackle other urgent problems to sustain the health of our planet.