Abstract:
Human induced aerosols, such as sulphate, cool the Earth by reflecting some sunlight back to space. They also change the development and lifecycle of clouds which, in turn, regulate aerosols. Our current inability to accurately quantify these complex effects undermines our ability to attribute historical trends and accurately predict future climate changes.
Here I will describe recent work using machine learning (ML) to tackle this challenge and improve our understanding of the effects of aerosol on our climate. Firstly, by describing the use of deep neural networks to assess the prevalence of isolated cloud perturbations in petabytes of satellite imagery, and causal models which allow us to unpick the effect of aerosol on clouds directly from such observations. Secondly, by introducing the use of model emulation and extensive aerosol measurements for the improved calibration of climate models. Finally, I will discuss the potential for energy-balance emulators to provide top-down constraints that more explicitly bridge the appropriate scales and improve our ability to model these important processes.