Machine Learning for Climate

The theoretical understanding of the Earth system has fundamentally advanced in recent decades in parallel to an exponential increase of observations and modeling data. However, climate scientists cannot meet the challenge of informing society about changes that may occur in the future at regional and local scales because many two-way, multi-scale processes that encompass the physical chemical and biological realms continue to elude us. Big data and the associated algorithms (Machine Learning) provide the opportunity to learn about quantities related to the climate systems in ways and with an amount of detail that were infeasible only a few years ago. The opportunity for descriptive inference creates the chance for climate scientists to ask causal questions and create new theories or validate old ones. Furthermore, when paired with modeling experiments or robust research in model parameterizations, “big data” can provide data-driven answers to vexing questions.

This conference will set the stage for exchanging tools and ideas and will help identify key problems where consistent progress is achievable through collaborative efforts. The theme of the conference will extend more broadly than the Physics focus of the main program, in order to elicit input from a wide range of experts across the earth system and computational sciences who are involved in the climate change problem. Given the level of interdisciplinarity and exchange that we aim for and expect, this conference will summarize current understanding and open questions, and will set the stage for achieving the aims of the associated KITP program.