2026 SPRING LECTURES IN CLIMATE DATA SCIENCE
January 22, 2026- May 7, 2026
- 12:00 – 1:30 pm (EDT)
- IN-PERSON at the Tang Family Hall (Rm 202) at the Columbia Engineering Innovation Hub (2276 12th Ave, New York, NY)
- VIRTUAL ATTENDANCE available
- Add to Calendar
Click to see past Lectures in Climate Data Science: Fall 2025, Summer 2025, Spring 2025, Fall 2024, Summer 2024, Spring 2024, Fall 2023, Spring 2023, and Fall 2022.
THURSDAY || JANUARY 22, 2026
SAMORY KPOTUFE
Columbia University
Theoretical works on supervised transfer learning (STL) — where the learner has access to labeled samples from both source and target distributions — have for the most part focused on statistical aspects of the problem, while efficient optimization has received less attention. We consider the problem of designing an SGD procedure for STL that alternates sampling between source and target data, while maintaining statistical transfer guarantees without prior knowledge of the quality of the source data. A main algorithmic difficulty is in understanding how to design such an adaptive sub-sampling mechanism at each SGD step, to automatically gain from the source when it is informative, or bias towards the target and avoid negative transfer when the source is less informative.
THURSDAY || FEBRUARY 5, 2026
Watch on YouTube
SAVANNAH FERRETTI
University of California, Irvine
Making Data-Driven Climate Prediction Interpretable by Design
Machine learning can improve climate prediction by combining information across horizontal space, height, and time in a “neighborhood” around each prediction point. However, as these neighborhoods grow, the learned relationships can become difficult to interpret and increasingly prone to overfitting.
We introduce data-driven integration kernels, a framework that builds interpretability into this setting by separating information aggregation from local nonlinear prediction. Each predictor field is first summarized by integrating it with learnable weighting functions (“kernels”) over space, height, and/or time, and the prediction model then operates only on these integrated features (plus any optional local inputs).
This structure limits complex nonlinear interactions while making each kernel directly interpretable as a weighting pattern, revealing which horizontal locations, vertical levels, and past timesteps contribute most to the prediction. We demonstrate the approach for South Asian monsoon rainfall, showing that kernel-based models achieve near-baseline performance with far fewer trainable parameters, suggesting that a small number of interpretable integrations can capture much of the relevant spatiotemporal information.
THURSDAY || FEBRUARY 19, 2026
Watch on YouTube
PETER DUEBEN
ECMWF
Generative Machine Learning to Change How We Do Earth System Modeling
Machine learning has a massive effect on numerical weather prediction and will likely also have a very strong effect on climate modeling. As a consequence, we are talking a lot about the new models and tools that we are working with in Earth system modeling. However, generative machine learning will also have a significant effect on how we work and how we build Earth system models in the future. This talk will try to outline future changes.
THURSDAY || FEBRUARY 26, 2026
DA FAN
Columbia University
Using Contrastive Learning to Identify Structural Deficiencies in Climate Models and Generate Calibrated Physics Ensembles
Calibrating climate models is challenging, labor-intensive, and time-consuming. To address this, we develop a neural network–based contrastive learning framework to accelerate the efforts aiming to auto-calibrate climate models. The contrastive learning model identifies structural biases in climate models. We also apply a Gaussian process model to learn the relationship between physical parameters and the differences between climate simulations and observations. Using this relationship, we generate calibrated parameter sets that improve agreement with observations.
THURSDAY || MARCH 5, 2026
NOAH BRENOWITZ
NVIDIA
Computational and Data Bottlenecks for Weather + Climate Modeling
Earth system models rely on the raw resources of data and computational power to produce high fidelity simulations. In this talk, I will begin by surveying some recent progress in data driven modeling of the atmosphere but quickly move onto to discuss where progress is being limited by a lack of resources. I will argue that physics models and AI approaches are naturally complementary. AI methods are intrinsically faster on modern hardware, but physics models need less data.
THURSDAY || MARCH 26, 2026
DHRUV BALWADA
Columbia University
Machine Learning Applications in Physical Oceanography: From Observations to Models
Ocean turbulence plays a central role in shaping the distribution of heat, carbon, and momentum across the climate system, yet representing its effects in models and quantifying its properties from observations remain open challenges. In this talk, I will present a series of projects that leverage machine learning to advance our understanding and representation of ocean turbulence across scales. First, I will discuss the development of artificial neural network-based mesoscale eddy parameterizations trained on high-resolution simulations and implemented within the MOM6 ocean model, highlighting both the promises and practical challenges of embedding learned closures in general circulation models. Second, I will describe ongoing work on parameterizing the effects of air-sea interaction heterogeneity driven by submesoscale and mesoscale ocean variability. Third, I will present efforts to apply modern generative modeling approaches to the assimilation of ocean surface fields, with a focus on using learned priors to constrain state estimation from sparse observations. Finally, I will discuss how data-driven methods can complement classical diagnostics of turbulence, particularly in characterizing cross-scale energy transfers at the submesoscale using structure function techniques applied to ocean observations. Together, these projects illustrate how machine learning can serve as a connective thread between observational analysis, theoretical understanding, and model development in physical oceanography.
THURSDAY || APRIL 9, 2026
NILS THÜEREY
Technische Universität München
Generative AI for Learned PDE Solving
n this talk I’ll explore the potential of generative AI techniques (such as denoising diffusion and flow matching) for learned PDE simulators. These models don’t just produce out a single best guess (the “mean”); they learn full probability distributions, letting us draw different samples and explore the range of possible outcomes. Hence, diffusion models transform learned representations into probabilistic models, enabling the generation of samples from complex posterior distributions rather than merely producing a deterministic estimate of the mean. Furthermore, integrating existing numerical methods into learning tasks allows for the creation of highly accurate inverse solvers. The ability of learned probabilistic surrogates to sample from the posterior is particularly promising for challenging downstream tasks, such as uncertainty quantification and improved interpretability.
THURSDAY || APRIL 16, 2026
BRUCE USHER
Columbia Business School
AI Growth, Power Demand, and the Implications for Climate Change
THURSDAY || APRIL 23, 2026
LEAP ML JOURNAL CLUB
(Jakhar, K., Guan, Y., and Hassanzadeh, P. (2026). Analytical and AI-Discovered Stable, Accurate, and Generalizable Subgrid-Scale Closure for Geophysical Turbulence. APS Physical Review Letters, vol. 136, no. 6, Art. no. 064201. doi:10.1103/v28b-5qmp.)
By combining artificial intelligence and fluid physics, we discover a closed-form closure for 2D turbulence from small direct numerical simulation data. Large-eddy simulation with this closure is accurate and stable, reproducing direct numerical simulation statistics, including those of extremes. We also show that the new closure could be derived from a fourth-order truncated Taylor expansion. Prior analytical and artificial-intelligence-based work only found the second-order expansion, which led to unstable large-eddy simulation. The additional terms emerge only when interscale energy transfer is considered alongside standard reconstruction criterion in the sparse-equation discovery.
THURSDAY || APRIL 30, 2026
JOSEPH KO
Columbia University
Compressing Complexity in Ice Clouds
Ice-containing clouds play a central role in Earth’s energy balance and hydrologic cycle, yet their microphysical representation remains a major source of uncertainty in atmospheric models. A key challenge is balancing physical realism with computational tractability. This talk explores how compression, via machine-learned low-dimensional representations, can provide a principled bridge between complexity and tractability for ice microphysical applications. First, I will discuss how self-supervised learning can be used to learn latent embeddings of ice crystal morphology from high-dimensional data (e.g., in situ images), and demonstrate how these embeddings enable new modes of analysis beyond traditional habit classifications. Second, I introduce latent diagnostics as compact summaries within the context of perturbed parameter ensembles, and show how they can reduce parametric uncertainty and improve parameter estimation. Together, these results highlight how compression can both enhance physical insight and facilitate improved parameterizations of ice clouds.
THURSDAY || MAY 7, 2026
HANG FAN
Columbia University
Efficient and Physically Consistent Atmospheric Data Assimilation in Latent Space
Data assimilation (DA) combines observations with numerical model forecasts to estimate atmospheric states. However, due to the high dimensionality of the atmosphere, traditional Bayesian approaches struggle with physical consistency, computational cost, and non-Gaussian processes. Here, we introduce Latent Data Assimilation (LDA), which performs DA in a learned latent space to overcome these challenges. This framework improves both efficiency and accuracy, while providing a more effective way to handle
nonlinearities in atmospheric systems.
THURSDAY || MAY 14, 2026
This Lecture has been postponed to July 9, 2026.
KARA LAMB
Columbia University