Environmental Science and Engineering Seminar

Cloud microphysical processes---those that make and evolve clouds and precipitation---are critical for understanding both how Earth's climate will respond to anthropogenic emissions as well as how high-impact weather systems such as thunderstorms evolve. However, these processes are challenging to tackle via data-driven methods, because there is no true reference model---cloud microphysical processes are uncertain at every scale. Data-driven approaches that leverage the formalism of Bayesian statistics and machine learning must therefore aim to improve reduced-order models with both the (flawed, incomplete) structurally complex models that do exist, as well as real observations. In many cases, errors are primarily structural in origin, rather than parametric, and we attempt to systematize the selection of reduced-order model structure. I will discuss efforts to improve cloud microphysics, within both high resolution large-eddy simulations, as well as for global Earth system models such as the NASA GISS ModelE3, the DOE's E3SM, and CESM. I will present recent breakthroughs in these efforts, and remaining challenges towards data-driven development and constraint at all scales.