Climate Analytics Lab

Improving climate projections with machine learning

These are optional background papers that go with the weekly topics. The required readings are on each week’s page. You don’t need to read all of these: skim the abstract and figures, and read further if a paper connects to something you’re working on. They are also a good place to start looking for Phase II ideas.

Weeks 1–2: CMIP and scenarios

  • Eyring, V., et al. (2016). Overview of the Coupled Model Intercomparison Project Phase 6 (CMIP6) experimental design and organization. Geoscientific Model Development 9, 1937–1958. doi:10.5194/gmd-9-1937-2016 Where the ClimateBench simulations come from, and why CMIP runs the experiments it does.
  • O’Neill, B. C., et al. (2016). The Scenario Model Intercomparison Project (ScenarioMIP) for CMIP6. Geoscientific Model Development 9, 3461–3482. doi:10.5194/gmd-9-3461-2016 What the SSP scenarios are, and how their emissions were designed.

Weeks 3–4: Emulators, and the model hierarchy

  • Lütjens, B., Ferrari, R., Watson-Parris, D., & Selin, N. E. (2025). The impact of internal variability on benchmarking deep learning climate emulators. Journal of Advances in Modeling Earth Systems 17. doi:10.1029/2024MS004619 Linear pattern scaling can outperform deep learning on ClimateBench once internal variability is accounted for. Read this before you trust your leaderboard.
  • Held, I. M. (2005). The gap between simulation and understanding in climate modeling. Bulletin of the American Meteorological Society 86, 1609–1614. doi:10.1175/BAMS-86-11-1609 A short essay on why intermediate-complexity models like SPEEDY are worth having.
  • Kochkov, D., et al. (2024). Neural general circulation models for weather and climate. Nature 632, 1060–1066. doi:10.1038/s41586-024-07744-y NeuralGCM: a hybrid ML–physics model built on the same Dinosaur dynamical core as JCM.

Weeks 5–6: Climate sensitivity, tuning, and differentiable models

  • Knutti, R., Rugenstein, M. A. A., & Hegerl, G. C. (2017). Beyond equilibrium climate sensitivity. Nature Geoscience 10, 727–736. doi:10.1038/ngeo3017 What ECS does and doesn’t tell us, and why feedbacks change as the planet warms.
  • Hourdin, F., et al. (2017). The art and science of climate model tuning. Bulletin of the American Meteorological Society 98, 589–602. doi:10.1175/BAMS-D-15-00135.1 How climate models are calibrated in practice, and the problem gradient-based calibration is trying to improve.
  • Gelbrecht, M., White, A., Bathiany, S., & Boers, N. (2023). Differentiable programming for Earth system modeling. Geoscientific Model Development 16, 3123–3135. doi:10.5194/gmd-16-3123-2023 The case for automatic differentiation in climate models, and its pitfalls.

Weeks 7–8: Emulating models, and simple climate models

  • Watson-Parris, D., Williams, A., Deaconu, L., & Stier, P. (2021). Model calibration using ESEm v1.1.0 – an open, scalable Earth system emulator. Geoscientific Model Development 14, 7659–7672. doi:10.5194/gmd-14-7659-2021 Emulating a climate model over its parameters, then using the emulator for calibration. This is the week 7 workflow.
  • Leach, N. J., et al. (2021). FaIRv2.0.0: a generalized impulse response model for climate uncertainty and future scenario exploration. Geoscientific Model Development 14, 3007–3036. doi:10.5194/gmd-14-3007-2021 The kind of simple climate model the IPCC uses for scenario assessment, and a natural global-mean component for a pattern-scaling hybrid.
  • Bonev, B., et al. (2023). Spherical Fourier neural operators: Learning stable dynamics on the sphere. Proceedings of the 40th International Conference on Machine Learning (ICML). arXiv:2306.03838 A neural-operator architecture designed for data on the sphere, and a candidate architecture for Phase II.

Weeks 9–10: Physical constraints, and ML for climate more broadly

  • Beucler, T., et al. (2021). Enforcing analytic constraints in neural networks emulating physical systems. Physical Review Letters 126, 098302. doi:10.1103/PhysRevLett.126.098302 Two ways to make a neural network conserve energy and mass, and what each costs.
  • Watson-Parris, D. (2021). Machine learning for weather and climate are worlds apart. Philosophical Transactions of the Royal Society A 379, 20200098. doi:10.1098/rsta.2020.0098 Why climate emulation is a harder, and different, problem from weather forecasting.