Coupling data science and numerical simulations to empower atmospheric and environmental research

We have entered the age of Data Science. Massive data from numerical simulations of the Earth system are now common in atmospheric and environmental research. However, state-of-the-art Earth System Models are subject to limitations because of the multiscale nature of the Earth system, where processes on scales smaller than the computational grid resolution remain unresolved and can only rely on simplified representations. These simplified representations introduce large yet frequently poorly characterized uncertainties in climate simulations. Therefore, making sense and making use of these simulation data remains a fundamental challenge. This dissertation tackles two critical representations in the Earth System Models: aerosol representation and urban representation. We couple Data Science and numerical simulations to create a suite of tools for addressing and overcoming the limitations: (1) Coarse graining & Regionalizing: Simulating a particle-resolved model to calculate the aerosol mixing state index (a metric to describe the chemical composition across the aerosol) at the global scale is computationally expensive. We develop data-driven emulators (surrogate models) that are learned from particle-resolved aerosol simulations to predict submicron aerosol mixing state indices from Earth System Model simulations. Different mixing state indices exhibit unique spatial and seasonal distributions via the global maps of aerosol mixing state indices. An unsupervised learning-based approach is developed to regionalize the global mixing state indices. We illustrate that regionalization can capture the variability in aerosol mixing state distribution. (2) Model benchmarking: Aerosols are often represented by several overlapping subpopulations with pre-defined parameters (known as “modes”) that cannot fully resolve mixing state. To quantitatively evaluate the error in mixing state represented by a modal model, we take the machine learning-enabled particle-resolved surrogate model as the benchmark model. The spatial patterns demonstrate the simplified aerosol representation assumption could induce large error (70%). (3) Uncertainty quantification: A detailed assessment of the uncertainty structure associated with urban heat wave projections on the global scale is critical but missing in the literature. An urban climate emulator framework is improved to project the global urban heat waves in the next several decades under climate change. We show that, at the urban scale a large proportion of the uncertainty results from choices of model parameter and structural design. These efforts culminate in an improved understanding of the role of Data Science in atmospheric and environmental research, in addition to the well-known importance of numerical simulations. The specific scientific outcomes of this work contribute to the current state of knowledge on atmospheric aerosols and urban environments. I hope this dissertation will inspire more innovations at the crossroad of Data Science and numerical simulations.

Name thesis:degree_name

Ph.D.

Level thesis:degree_level

Dissertation

Discipline thesis:degree_discipline

Environ Engr in Civil Engr

Grantor

University of Illinois at Urbana-Champaign

Year dc:date

2021

Author dc:creator

• Zheng, Zhonghua

Contributors dc:contributor

• Riemer, Nicole
• Zhao, Lei
• West, Matthew
• Anantharaj, Valentine G.

• Data Science
• Atmospheric Aerosols
• Urban Environments

Statement dc:rights

• Copyright 2020 Zhonghua Zheng

Language dc:language

en

Handle dc:identifier

http://hdl.handle.net/2142/109594

OAI identifier oai:identifier

oai:www.ideals.illinois.edu:2142/109594