Can We Reduce Structural and Parametric Uncertainties in Models of Cloud Microphysics?
Emily de Jong, California Institute of Technology
Droplets, aerosols, and ice particles in the atmosphere exist and interact at the microscale, but collectively they make up clouds, which control precipitation and amplify global warming through radiative feedbacks. Global models of the atmosphere operate at resolutions as low as tens or hundreds of kilometers, leaving a gap of several orders of magnitude between the scales of microphysics and resolved quantities. Current approaches to modeling microphysics involve aggregating properties of the particle populations into bulk quantities to make them computationally tractable at the global scale. However, these simplifications introduce errors and uncertainty through both the structure of the models, and through uncertainty in physical parameters which are not directly observable.
This poster outlines a two-pronged approach to reducing microphysics uncertainties based on recent and ongoing research. To attack structural uncertainties, we developed a generalized linear spectral method to track a size distribution of a droplet population. Current work aims to extend this method to a nonlinear approach based on the idea of “moving basis functions,” and preliminary evidence shows that this hybrid technique of treating droplet populations as subdistributions could reduce the computational burden of spectral methods. In conjunction with these improvements, we describe a set of recent studies using ensemble learning methods to solve for microphysics process parameters as an inverse problem. Finally, we extend this philosophy of learning low-order microphysics parameters from high-resolution simulations by outlining ongoing efforts to model microphysics thermodynamic tendencies directly through Gaussian process regression.
Abstract Author(s): Emily de Jong