AI4ClimateScience

Bridge Program at AAAI 2023 - February 2023

Bridge Overview

The bridge will overview the role of AI in modeling various aspects of climate – atmospheric, oceanic, and land processes. For each of these topics, we will include discussions on the available data sources, the currently deployed methods, and (if any) AI approaches that have been explored in prior works. A key part of these discussions will be the limitations on both the data and modeling ends and potential solutions. Some of these questions include:

1. What are the unique biases persisting existing datasets for climate science?
2. How can we best incorporate physical domain knowledge in AI approaches?
3. How can we model multimodal climate data spanning satellite imagery, simulation data, weather balloons, etc.?
4. How robust are machine learning models for weather and climate across space and time?
5. How can we quantify and calibrate uncertainties for forecasts with machine learning models?
6. How can we scale AI infrastructure for training and inference to the petabyte-scale climate science datasets available today?

Bridge Objectives

Climate change has led to a rapid increase in the occurrence of extreme weather events globally, including floods, droughts, and heatwaves. Modeling climate is an essential endeavor to understand the near and long term impacts of climate change, as well as inform tools for mitigation and adaptation. However, first-principles models are limited in their ability to model complex dynamics that dictate weather and climate phenomena. With advancements in sensory technology and machine learning, AI can play a key role in advancing data-driven climate modeling with accurate and high-resolution forecasts of a variety of spatiotemporal phenomena related to weather and climate, such as agriculture yields, natural disasters, and socioeconomic welfare. In this bridge program, we aim to bring together students, researchers, and practitioners in both AI and climate science on topics related to interdisciplinary research and education in applying machine learning methods for climate science. Through an extensive program consisting of hands-on tutorials, paper presentations, and invited talks and panels with domain experts, we aim to familiarize the audience with the most pressing questions in climate science that can benefit from AI, the limitations of current data-driven models (e.g., incorporating physical constraints, quantifying uncertainty, distribution shifts), as well as educational resources and community initiatives for students, researchers, and practitioners in both disciplines.

Acknowledgements

Previous version of this CFP contained content from the AAAI bridge on Continual Causality. Please check out their page at: https://www.continualcausality.org/.