Journal Article
Geoscientific Model Development, vol. 12, iss. 7, pp. 2679-2706, 2019
Authors
Qi Tang, Stephen A. Klein, Shaocheng Xie, Wuyin Lin, Jean-Christophe Golaz, Erika L. Roesler, Mark A. Taylor, Philip J. Rasch, David C. Bader, Larry K. Berg, Peter Caldwell, Scott E. Giangrande, Richard B. Neale, Yun Qian, Laura D. Riihimaki, Charles S. Zender, Yuying Zhang, Xue Zheng
Abstract
Abstract. Climate simulations with more accurate process-level
representation at finer resolutions (<100 km) are a pressing need in
order to provide more detailed actionable information to policy makers
regarding extreme events in a changing climate. Computational limitation is
a major obstacle for building and running high-resolution (HR, here
0.25∘ average grid spacing at the Equator) models (HRMs). A more
affordable path to HRMs is to use a global regionally refined model (RRM),
which only simulates a portion of the globe at HR while the remaining is at
low resolution (LR, 1∘). In this study, we compare the Energy Exascale
Earth System Model (E3SM) atmosphere model version 1 (EAMv1) RRM with the HR
mesh over the contiguous United States (CONUS) to its corresponding globally
uniform LR and HR configurations as well as to observations and reanalysis
data. The RRM has a significantly reduced computational cost (roughly
proportional to the HR mesh size) relative to the globally uniform HRM. Over
the CONUS, we evaluate the simulation of important dynamical and physical
quantities as well as various precipitation measures. Differences between
the RRM and HRM over the HR region are predominantly small, demonstrating
that the RRM reproduces the precipitation metrics of the HRM over the CONUS.
Further analysis based on RRM simulations with the LR vs. HR model
parameters reveals that RRM performance is greatly influenced by the
different parameter choices used in the LR and HR EAMv1. This is a result of
the poor scale-aware behavior of physical parameterizations, especially for
variables influencing sub-grid-scale physical processes. RRMs can serve as a
useful framework to test physics schemes across a range of scales, leading
to improved consistency in future E3SM versions. Applying
nudging-to-observations techniques within the RRM framework also
demonstrates significant advantages over a free-running configuration for
use as a test bed and as such represents an efficient and more robust
physics test bed capability. Our results provide additional confirmatory
evidence that the RRM is an efficient and effective test bed for HRM
development.
