GISS Lunch Seminar

Speaker: Kara Lamb (Columbia Univ.)
Topic: Learning Cloud Processes Across Scales Using Data-Driven Reduced Order Modeling

Clouds are challenging to represent in climate models because they involve complex, non-linear processes that extend from the submicron scale to the kilometer scale. Recent data-driven approaches developed in the context of scientific machine learning are promising methods for deriving consistent representations of cloud processes across different scales. I will talk about a couple of recent studies applying these methods to cloud processes, both at the microscale and at the macroscale. Representing cloud microphysical processes in large scale atmospheric models is challenging because many processes depend on the details of the droplet size distribution (the spectrum of droplets with different sizes in a cloud). I will discuss how data-driven reduced order modeling can be used to learn predictors for microphysical process rates in bulk microphysics schemes in an unsupervised manner from higher dimensional bin distributions, by simultaneously learning lower dimensional representations of droplet size distributions and predicting the evolution of the microphysical state of the system. I will also discuss how these methods can be used to parameterize cloud spatial organization at the mesoscale in climate models, using simulations from a global storm resolving model that resolves deep convection. Learning a parameter for cloud spatial organization significantly improves the prediction of the precipitation in climate models, and almost entirely explains precipitation stochasticity at the sub-grid-scale.