2025 FALL LECTURES IN CLIMATE DATA SCIENCE
September 11, 2025 - December 18, 2025
- 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: Summer 2025, Spring 2025, Fall 2024, Summer 2024, Spring 2024, Fall 2023, Spring 2023, and Fall 2022.
THURSDAY || SEPTEMBER 4, 2025
LEAP ML JOURNAL CLUB
“The Impact of Internal Variability on Benchmarking Deep Learning Climate Emulators”
(, , , & (2025). The impact of internal variability on benchmarking deep learning climate emulators. Journal of Advances in Modeling Earth Systems, 17, e2024MS004619. https://doi.org/10.1029/2024MS004619.)
Full-complexity Earth system models (ESMs) are computationally very expensive, limiting their use in exploring the climate outcomes of multiple emission pathways. More efficient emulators that approximate ESMs can directly map emissions onto climate outcomes, and benchmarks are being used to evaluate their accuracy on standardized tasks and data sets. We investigate a popular benchmark in data-driven climate emulation, ClimateBench, on which deep learning-based emulators are currently achieving the best performance. We compare these deep learning emulators with a linear regression-based emulator, akin to pattern scaling, and show that it outperforms the incumbent 100M-parameter deep learning foundation model, ClimaX, on 3 out of 4 regionally resolved climate variables, notably surface temperature and precipitation. While emulating surface temperature is expected to be predominantly linear, this result is surprising for emulating precipitation. Precipitation is a much more noisy variable, and we show that deep learning emulators can overfit to internal variability noise at low frequencies, degrading their performance in comparison to a linear emulator. We address the issue of overfitting by increasing the number of climate simulations per emission pathway (from 3 to 50) and updating the benchmark targets with the respective ensemble averages from the MPI-ESM1.2-LR model. Using the new targets, we show that linear pattern scaling continues to be more accurate on temperature, but can be outperformed by a deep learning-based technique for emulating precipitation. We publish our code and data at https://github.com/blutjens/climate-emulator.
FRIDAY || SEPTEMBER 5, 2025
THIS LECTURE HAS BEEN POSTPONED TO October 30, 2025.
STEPHAN MANDT
UC Irvine
THURSDAY || SEPTEMBER 11, 2025
THIS LECTURE HAS BEEN POSTPONED TO APRIL 30, 2026.
CE BIAN
Lamont-Doherty Earth Observatory / Columbia University
THURSDAY || SEPTEMBER 18, 2025
DAVID HOGG
NYU
Is Machine Learning Good or Bad for the Natural Sciences?
Machine learning (ML) methods are having a huge impact across all of the sciences. However, ML has a strong ontology — in which only the data exist — and a strong epistemology — in which a model is considered good if it performs well on held-out training data. These philosophies are in strong conflict with both the practice and the philosophy of the natural sciences. I identify some locations for ML in the natural sciences at which the ontology and epistemology are valuable. I also show that there are contexts in which the introduction of ML introduces strong, unwanted statistical biases. I will especially make comments about the use of ML for emulation and surrogate models, and the use of ML to make new (labeled, say) datasets.
TUESDAY || SEPTEMBER 23, 2025
JOAO TEIXEIRA
NASA Jet Propulsion Laboratory / California Institute of Technology
Turbulence, Convection + Climate: Global Atmospheric Models and a Future Satellite Mission
Atmospheric boundary layer turbulence and convection play a critical role in our weather and climate. However, turbulence and convection processes are notoriously difficult to model and observe in an adequate manner, and several key science problems remain unsolved. In this presentation, the critical issue of how to represent turbulent and convective mixing in a realistic manner in weather and climate prediction models is addressed from an integrated perspective. Recent climate model results highlight the potential of merging complementary theories on how to represent turbulence and convection. In this context, a future satellite mission to better observe the thermodynamic structure of the atmospheric boundary layer from space will be discussed.
THURSDAY || SEPTEMBER 25, 2025
** This event is part of Climate Week at Columbia Engineering. **
“GEN-AI FOR CLIMATE SCIENCE”
LEAP hosts a timely panel discussion on the transformative role of Generative AI (GenAI) in advancing climate science. As large-scale data and complex modeling demands grow across the climate research community, GenAI methods offer new possibilities for emulation, downscaling, prediction, and decision support. Join us for a rich discussion about how GenAI is accelerating progress in climate science.
This interdisciplinary panel brings together leading experts from academia, industry, and technology to explore how GenAI is reshaping key areas of climate science—from ocean modeling and data assimilation to climate emulation and real-time chatbot interfaces. Panelists will share insights on current applications, technical challenges, and future directions for integrating GenAI tools in research and operational contexts.
Panelists:
- Opening Remarks, by Pierre Gentine (Maurice Ewing and J. Lamar Worzel Professor of Geophysics and Professor of Earth and Environmental Sciences, Columbia University; Professor at Columbia Climate School; Director, LEAP)
- “AI-enabled Ocean Modeling,” by Laure Zanna (Joseph B. Keller and Herbert B. Keller Professor in Applied Mathematics; Professor of Mathematics + Data Science, NYU; Geoscience Director, LEAP)
- “GenAI for Climate Projection,” by Francesco Immorlano (Postdoctoral Researcher, University of California Irvine)
- “GenAI for Steerable Climate Informatics: Climate in a Bottle,” by Mike Pritchard (Director of Climate Simulation Research, NVIDIA; Professor of Earth System Science, UC Irvine)
- Additional Remarks, by George Deodatis (Vice Dean for Research, School of Engineering and Applied Science; Santiago and Robertina Calatrava Family Professor of Civil Engineering and Engineering Mechanics; Professor of Earth and Environmental Engineering, Columbia University)
- “GenAI for Data Assimilation,” by Gideon Dresdner (Research Engineer, Brightband)
- “GenAI for Downscaling,” by Ignacio Lopez Gomez (Research Scientist, Google Research)
- “A Vision for Agentic AI for Climate Research,”
by Tian Zheng (Professor and Department Chair, Department of Statistics, Columbia University; Deputy Director, LEAP)
THURSDAY || OCTOBER 2, 2025
“LatticeVision: Image to Image Networks for Modeling Non-Stationary Spatial Data”
(Sikorski, A., Ivanitskiy, M., Lenssen, N., Nychka, D., & McKenzie, D.(2025). LatticeVision: Image to Image Networks for Modeling Non-Stationary Spatial Data. arXiv preprint arXiv:2505.09803.)
In many scientific and industrial applications, we are given a handful of instances (a ‘small ensemble’) of a spatially distributed quantity (a ‘field’) but would like to acquire many more. For example, a large ensemble of global temperature sensitivity fields from a climate model can help farmers, insurers, and governments plan appropriately. When acquiring more data is prohibitively expensive — as is the case with climate models — statistical emulation offers an efficient alternative for simulating synthetic yet realistic fields. However, parameter inference using maximum likelihood estimation (MLE) is computationally prohibitive, especially for large, non-stationary fields. Thus, many recent works train neural networks to estimate parameters given spatial fields as input, sidestepping MLE completely. In this work we focus on a popular class of parametric, spatially autoregressive (SAR) models. We make a simple yet impactful observation; because the SAR parameters can be arranged on a regular grid, both inputs (spatial fields) and outputs (model parameters) can be viewed as images. Using this insight, we demonstrate that image-to-image (I2I) networks enable faster and more accurate parameter estimation for a class of non-stationary SAR models with unprecedented complexity.
THURSDAY || OCTOBER 16, 2025
THIS LECTURE HAS BEEN POSTPONED TO January 22, 2026.
SAMORY KPOTUFE
Columbia University
THURSDAY || OCTOBER 23, 2025
SALVA RÜHLING CACHAY
UC San Diego
Efficient Generative Modeling for Weather and Climate Prediction
Accurate and reliable weather and climate prediction is vital for societal planning, disaster mitigation, and understanding climate change. The increasing frequency of extreme weather events demands robust probabilistic forecasts that effectively quantify uncertainty. While generative models, particularly diffusion models, offer a promising path, methodological innovations are needed to overcome crucial limitations in their computational and data efficiency for these large-scale problems. In this talk, I will discuss methodological advancements that address these efficiency bottlenecks by explicitly leveraging the temporal structure of the data to design more efficient generative models for forecasting and emulation problems.
THURSDAY || OCTOBER 30, 2025
STEPHAN MANDT
UC Irvine
Scientific Inference with Diffusion Generative Models
Diffusion models have revolutionized generative modeling in vision and language. But can they also serve as tools for scientific inference? I present a Bayesian view on diffusion models as tools for solving inverse problems—inferring hidden states or parameters from noisy observational data—with applications ranging from climate modeling to imaging. Unlike creative applications, scientific use cases demand more than photorealism: they require calibrated uncertainty, faithful distributions, efficient conditional sampling, and the capacity to represent extreme and heavy-tailed data. I will highlight four recent advances addressing these needs: Variational Control for expressive conditional generation (ICML ’25), Heavy-Tailed Diffusion Models for sparse and extreme-valued data (ICLR ’25), Conjugate Integrators for fast conditional sampling without retraining (NeurIPS ’24), and Generative Uncertainty for assessing epistemic uncertainty in generation (UAI ’25). Together, these results point to diffusion models as principled engines of scientific inference.
THURSDAY || NOVEMBER 6, 2025
“Embedding Machine-Learnt Sub-Grid Variability Improves Climate Model Precipitation Patterns”
(Giles, D., Briant, J., Morcrette, C.J. et al. Embedding machine-learnt sub-grid variability improves climate model precipitation patterns. Commun Earth Environ 5, 712 (2024). https://doi.org/10.1038/s43247-024-01885-8.)
Full-complexity Earth system models (ESMs) are computationally very expensive, limiting their use in exploring the climate outcomes of multiple emission pathways. More efficient emulators that approximate ESMs can directly map emissions onto climate outcomes, and benchmarks are being used to evaluate their accuracy on standardized tasks and data sets. We investigate a popular benchmark in data-driven climate emulation, ClimateBench, on which deep learning-based emulators are currently achieving the best performance. We compare these deep learning emulators with a linear regression-based emulator, akin to pattern scaling, and show that it outperforms the incumbent 100M-parameter deep learning foundation model, ClimaX, on 3 out of 4 regionally resolved climate variables, notably surface temperature and precipitation. While emulating surface temperature is expected to be predominantly linear, this result is surprising for emulating precipitation. Precipitation is a much more noisy variable, and we show that deep learning emulators can overfit to internal variability noise at low frequencies, degrading their performance in comparison to a linear emulator. We address the issue of overfitting by increasing the number of climate simulations per emission pathway (from 3 to 50) and updating the benchmark targets with the respective ensemble averages from the MPI-ESM1.2-LR model. Using the new targets, we show that linear pattern scaling continues to be more accurate on temperature, but can be outperformed by a deep learning-based technique for emulating precipitation. We publish our code and data at https://github.com/blutjens/climate-emulator.
THURSDAY || NOVEMBER 13, 2025
ZHOU YU
Columbia University
From POC to Profit: Leveraging Synthetic Users for Automated Testing + Evaluation
95% of the enterprise AI agents are stuck in POC (proof-of-concept). We don’t know when agents are good enough to deploy without systematic testing and evaluation process. We’ll share how to leverage AI to automate this process. In addition, we’ll talk about how to improve agents and make them self-learn through reinforcement learning (RL).
THURSDAY || NOVEMBER 20, 2025
MARY MARGARET STOLL
University of Washington
“Leveraging 4-Dimensionally Mapped Ocean Biogeochemistry to Inform Species Distribution Modeling”
Understanding the drivers of animal migratory patterns in a four-dimensional (4D) ocean is essential for developing reliable species distribution models (SDMs) that support ecosystem-based management. Despite recent advances in biologging and monitoring technologies, many SDMs overlook dynamic, subsurface biogeochemical conditions that impose critical constraints on suitable habitat. The Biogeochemical Argo (BGC-Argo) float array has revolutionized our view of the 4D biogeochemical landscape, providing new perspectives on the linkages between species distributions and environmental conditions. To assess the utility of depth-resolved, biogeochemical information in SDMs, we leverage archival tagging datasets, mapped ocean biogeochemistry data products, and machine learning methods to examine the extensive migrations of North Pacific albacore tuna (Thunnus alalunga), a commercially important fishery species. Our SDM predicts habitat suitability with high skill across space and time using multiple 4D biogeochemical predictor variables that capture both environmental deterrents (e.g. oxygen limitation) and attractants (e.g. chlorophyll, particulate organic carbon). These conditions strongly impact albacore vertical distributions, influencing both seasonal shifts and large-scale migration patterns. While temperature is a key determinant of habitat suitability, our findings demonstrate that biogeochemical variables substantially expand or contract habitat beyond what is thermally suitable. In this study, biogeochemical predictors contract suitable habitat by more than 60% in winter and expand it by 25% in summer, emphasizing the influence of subsurface biogeochemistry in modulating habitat availability. This scalable framework can be applied to other taxa and regions, offering new avenues to improve bycatch avoidance, catch per unit effort estimates, and predictions of climate-driven habitat shifts.
THURSDAY || NOVEMBER 20, 2025
SAM SILVA
University of Southern California
Data Science, AI, and Atmospheric Composition
Understanding the controlling factors behind the chemical composition of the Earth’s atmosphere is a critical step toward addressing the modern environmental challenges of air pollution and climate change. Traditional methods interrogating theoretical predictions with observations have been highly successful in addressing these challenges, particularly in light of the recent immense growth of data availability in environmental systems. However, there are still gaps in our scientific knowledge due to limitations in modern scientific techniques (e.g., theoretical frameworks, observational systems, and computational power). Data-driven methods from the data science and artificial intelligence literature, when informed and guided by scientific understanding, present a valuable tool in addressing these knowledge gaps. In this seminar, I will present results from recent work using a variety of data science and A.I. methods to better constrain modern understanding of atmospheric chemistry and composition.
THURSDAY || DECEMBER 11, 2025
HAMIDREZA NOROUZI
New York City College of Technology
Remote Sensing of Lakes, Urban Climate, and Heat Resilience from Local to Global
This talk presents an integrated overview of our remote sensing and GeoAI research spanning lake ecosystems, urban heat, and subsurface transit environments. Using multi-sensor satellite data and in-situ measurements, we quantify trends in lake surface temperature, ice phenology, and harmful algal blooms across Adirondack and global
lakes, producing high-resolution datasets that support long-term ecological monitoring.
In urban settings, we combine downscaled land surface temperatures, micro-meteorological observations, and machine-learning models to map neighborhood-scale heat exposure and develop Cool Routing pathways that reduce pedestrian thermal stress.
Findings from a comprehensive NYC subway climate study are additionally presented, drawing on platform sensors, infrared imaging, and forecasting models to estimate station-level temperature and heat index up to ten days in advance.
Together, these efforts demonstrate how remote sensing and AI-driven analytics can create actionable climate services that enhance environmental resilience, public health, and community decision-making.
THURSDAY || DECEMBER 18, 2025
THIS LECTURE HAS BEEN POSTPONED TO A LATER DATE.
BO LI
Washington University in St. Louis
Spatial Changepoint Detection to Map the Impact of the Mount Pinatubo Eruption
Significant events such as volcanic eruptions can exert global and long-lasting impacts on climate. These impacts, however, are not uniform across space and time. Motivated by the need to understand how the 1991 Mt. Pinatubo eruption influenced global and regional climate, we propose a Bayesian framework to simultaneously detect and estimate spatially varying temporal changepoints. Our approach accounts for the diffusive nature of volcanic effects and leverages spatial correlation. We then extend the changepoint detection problem to large-scale spherical spatiotemporal data and develop a scalable method for global applications. The framework enables Gibbs sampling for changepoints within MCMC, offering greater computational efficiency than the Metropolis–Hastings algorithm. In addition, we incorporate spherical harmonic transformations to address the high dimensionality of global data, which further substantially reduce computational burden while preserving accuracy. We demonstrate the effectiveness of our method using both simulated datasets and real data of stratospheric aerosol optical depth and surface temperature to detect and estimate changepoints associated with the Mt. Pinatubo eruption.