Speaker: Associate Professor Pierre Gentine from Columbia University

 Title: "Machine learning to improve carbon/climate feedbacks"

Abstract: We are witnessing profound changes in our observational capacity of the biosphere and atmosphere with new generations of remote sensing platforms. Simultaneously, advances in our computational capacity have allowed us to better resolve multiple scales of the atmosphere from molecular scale to deep convective clouds. Yet, we still have huge uncertainties in two climate processes:

1) The capacity of continents to take up more CO2 in the future, which impact the CO2 concentration for a given emission rate.

2) The response of climate to prescribed CO2 concentration, which is mostly due to uncertainties in cloud processes.

In this talk, I will show how machine learning and physical approaches can be combined to address some of those questions. I will in particular show how these techniques can be used to better quantify global environmental changes in cold regions and their impact on carbon uptake, using remote sensing data and in situ observations. Then I will show how machine learning and high-resolution simulations could potentially be used to resolve the so-called cloud convection “deadlock", which is at the core of uncertainties in our climate predictions. [related papers 1, 2 & 3