Improved Climate Modeling through Machine-learning and Data-driven Approaches

Paul O’Gorman, Associate Professor of Earth, Atmospheric & Planetary Sciences
Elfatih Eltahir, Breene M. Kerr Professor of Civil and Environmental Engineering

Proposed Work

Climate models are a key tool for future projections of climate change and the associated impacts on society and ecosystems. However, climate models exhibit regional biases and uncertain feedback processes that limit the accuracy of climate projections. An important contributor to these biases and uncertainties is the representation of unresolved processes in the atmosphere, ocean and land surface through semi-empirical subgrid models known as parameterizations. New high-resolution modeling and observational datasets provide an unprecedented opportunity to greatly improve the parameterizations used in climate models. Here we propose to develop a new class of parameterizations for the atmosphere and ocean by combining machine learning algorithms with high-resolution simulations, and we propose to better constrain climate models over land using observational data. The overall goal of the proposed research is to investigate whether machine-learning and data-driven approaches may be used to improve climate models so that they are more useful for projections of climate change and for scientific investigations of the climate system.