Combining Geophysical Models With Machine Learning to Improve our Understanding of Climate Change

Chris Fletcher is a climate scientist and contributing author to the sixth IPCC assessment report.

Dr Fletcher’s research group at the University of Waterloo focuses on improving the development, calibration, and accuracy of Earth system models using a combination of satellite remote sensing and machine learning.

A particular area of focus is on monitoring and future projections of climate change in cold regions, answering science questions like “How much snow is on the ground in the Canadian Arctic?”, and “How is the snow pack going to change in a warmer world?”.

When not in front of a computer, Chris can usually be found somewhere on a bike.

Join us for this event with Dr. Fletcher hosted by MS2Discovery at Wilfrid Laurier University.

Title: Combining Geophysical Models With Machine Learning to Improve our Understanding of Climate Change.

Date and Time:  November 15, 2022 | 3 pm (EST)

Location:  LH3058 (Lazaridis Hall, Room 3058)

Decision-making and adaptation to climate change requires quantitative projections of the physical climate system and an accurate understanding of the uncertainty in those projections. Earth system models (ESMs) are numerical (physics-based) computational models that solve the Navier-Stokes equations on the sphere. ESMs provide the only reliable tool that climate scientists have to make projections forward into climate states that have not been observed in the historical data record. However, ESMs are computationally expensive and occupy the world's largest supercomputers, putting limits on their spatial resolution and/or physical accuracy. In contrast, data-driven models used in machine learning are relatively inexpensive to run, but their accuracy is limited by the quality and quantity of available training data. In this talk, I will present examples of projects from my research group that combine geophysical models with machine learning (ML) to target improvements in the efficiency, resolution and accuracy of ESMs. Examples I will present include using a deep neural network to learn the behaviours of a higher-resolution ESM using training data from lower-resolution (and thus computationally inexpensive) versions of the same ESM; a novel bias-correction technique for high-resolution estimates of snow in Ontario; and, a deep learning model to improve estimation of snow accumulation from radar measurements. Collectively, this research demonstrates that significant value can be added by combining data-driven models and geophysical models for problems spanning a range of temporal and spatial scales.