A Unified Data-Informed Model of Turbulence and Convection for Climate Prediction

Citation

López Gómez, Ignacio (2023) A Unified Data-Informed Model of Turbulence and Convection for Climate Prediction. Dissertation (Ph.D.), California Institute of Technology. doi:10.7907/042m-9686. https://resolver.caltech.edu/CaltechTHESIS:11152022-215747755

Abstract

Resolving atmospheric turbulent and convective processes in global climate simulations is, and will remain for decades, an intractable computational problem. The strong influence of these processes on cloud formation and maintenance makes the task of modeling turbulence and convection one of the grand challenges in climate modeling, due to the outsized effect of clouds on climate. Current operational climate models fail to represent atmospheric turbulence and convection accurately and consistently across dynamical regimes and vertical levels; errors in the representation of these processes explain about half of the spread in climate projections. This dissertation seeks to reduce such representation errors by improving a recently proposed unified framework for modeling turbulence and convection, known as the extended eddy-diffusivity mass-flux scheme, in several ways. First, the framework is rederived by systematically coarse-graining the governing fluid equations, highlighting the assumptions about atmospheric motion that are necessary to yield the scheme. New terms related to turbulent entrainment processes are shown to arise from the derivation. Second, a generalized formulation of turbulent diffusion consistent with the framework is presented. This novel formulation is shown to accurately represent turbulent processes under statically stable and unstable conditions, including regimes with sharp lapse rate inversions such as the stratocumulus-topped boundary layer. Finally, a methodology to calibrate free parameters within the model from indirect data is proposed. The methodology, based on Kalman filtering, is shown to be efficient at calibrating imperfect black-box models from noisy data, and in its regularized unscented version approximately quantifies parametric uncertainty. The resulting unified data-informed model of turbulence and convection is shown to accurately represent a range of low-cloud regimes that are associated with the largest biases in current operational climate models. The response of the model to realistic climate perturbations is also shown to be consistent with the resolved climate response, although structural errors in the amount of condensate are still important at realistic vertical resolutions.

Thesis Files