Julia Kaltenborn on the Juneau Icefield
Juneau Icefield Research Program, Camp 10, 2023 - learning from nature in nature.

Julia Kaltenborn

she/her

PhD Candidate in Computer Science at McGill University & Mila - Quebec AI Institute

Advised by Prof. David Rolnick

I am a scientific machine learning researcher developing methods that enable us to ask new scientific questions about the Earth system, with a particular focus on the cryosphere.

I care deeply about scientific problems in Earth system and cryospheric sciences and develop machine-learning methods around the constraints they impose. Many unsolved challenges in these fields require working with knowledge that already exists but is fragmented, partial, or expensive to obtain. Machine learning can make such knowledge more accessible by integrating different sources and learning representations that allow us to reason with them in new ways. However, when learned representations themselves become part of how we study partially observed systems, we also need ways to scientifically interrogate the knowledge they encode.

Selected Research

Emulating Numerical Models

My work on emulators focuses on future climate projections, using machine learning to make knowledge encoded in expensive numerical models more accessible across different Earth system models and forcing scenarios.

ClimateSet: A Large-Scale Climate Model Dataset for Machine Learning
Julia Kaltenborn, Charlotte E. E. Lange , et al. · NeurIPS 2023
Emulating the Forced Response of Climate Models with Generative Machine Learning
Graham Clyne, Julia Kaltenborn , et al. · 2026 · under review

Interrogating ML Models

My work explores causal representation learning as a path toward scientifically interrogating learned models of Earth systems, for example by enabling counterfactual questions about what they have learned.

Learning a Spatial Partitioning and its Causal Relations from Temporal Data
Philippe Brouillard, Sébastien Lachapelle, Julia Kaltenborn , et al. · CLeaR 2026
Causal Climate Emulation with Bayesian Filtering
Sebastian H. M. Hickman, Ilija Trajković, Julia Kaltenborn , et al. · NeurIPS 2025

Research Philosophy

To me, researching is a deeply human act. It emerges when we follow our curiosity - playfully, collaboratively, and carefully - question our own beliefs, and cross boundaries into other disciplines or into the unknown. As a scientist, I consider it my duty to question my own privileges and our complicity in extractive systems, particularly in contemporary AI research. In my computational work, I want to remain grounded in the physical systems, people, and practices through which data are produced. In practice, this means: going out into the field.

Community

My research would not be possible without the scientific, field, activist, and local communities that have raised and nurtured me. I try to pass some of that support on by initiating scientific communities (ML4Cryo), co-organizing humanitarian initiatives (AI Helps Ukraine Charity Conference), and creating interdisciplinary spaces where different ways of knowing can meet (Polar Eclipse School).

Selected Honours