2022
DOI: 10.1029/2021ms002794
|View full text |Cite
|
Sign up to set email alerts
|

Correcting Coarse‐Grid Weather and Climate Models by Machine Learning From Global Storm‐Resolving Simulations

Abstract: Global atmospheric “storm‐resolving” models with horizontal grid spacing of less than 5 km resolve deep cumulus convection and flow in complex terrain. They promise to be reference models that could be used to improve computationally affordable coarse‐grid global climate models across a range of climates, reducing uncertainties in regional precipitation and temperature trends. Here, machine learning of nudging tendencies as functions of column state is used to correct the physical parameterization tendencies o… Show more

Help me understand this report
View preprint versions

Search citation statements

Order By: Relevance

Paper Sections

Select...
2
1
1
1

Citation Types

0
50
0

Year Published

2022
2022
2022
2022

Publication Types

Select...
5
2
1

Relationship

2
6

Authors

Journals

citations
Cited by 48 publications
(57 citation statements)
references
References 25 publications
0
50
0
Order By: Relevance
“…The model calibrated using thermodynamic profiles improves upon the prior model in the forecast of horizontal velocities within the boundary and cloud layers. A common reason to use tendencies for calibration is the use of supervised learning techniques that are easy to implement for neural network architectures (e.g., Bretherton et al, 2022). In the next subsection, we demonstrate the power of UKI and EKI to calibrate hybrid models with embedded neural network parameterizations.…”
Section: The Identifiability Of Individual Parameters As a Function O...mentioning
confidence: 96%
“…The model calibrated using thermodynamic profiles improves upon the prior model in the forecast of horizontal velocities within the boundary and cloud layers. A common reason to use tendencies for calibration is the use of supervised learning techniques that are easy to implement for neural network architectures (e.g., Bretherton et al, 2022). In the next subsection, we demonstrate the power of UKI and EKI to calibrate hybrid models with embedded neural network parameterizations.…”
Section: The Identifiability Of Individual Parameters As a Function O...mentioning
confidence: 96%
“…The use of partial observations also highlights the benefits of learning from time statistics instead of tendencies. Learning from statistics not only ensures that the calibrated dynamical model is stable, which requires a leap of faith when training on instantaneous tendencies (Bretherton et al., 2022). It also couples the evolution of thermodynamic and dynamical fields, which can improve the forecast of fields unseen during training.…”
Section: Application To An Atmospheric Subgrid‐scale Modelmentioning
confidence: 99%
“…The model calibrated using thermodynamic profiles improves upon the prior model in the forecast of horizontal velocities within the boundary and cloud layers. A common reason to use tendencies for calibration is that they enable the use of supervised learning techniques, which are easy to implement for neural network architectures (Bretherton et al., 2022). In the next subsection, we demonstrate the power of UKI and EKI to calibrate hybrid models with embedded neural network parameterizations.…”
Section: Application To An Atmospheric Subgrid‐scale Modelmentioning
confidence: 99%
“…or hydrology [6]. The majority of ML-based parametrizations, however, is local [48,94,17,18,19,141,26,6,54,79,105,78,99,136,110], i.e., the in-and output are variables of single grid points, which assumes perfect scale separation, for example, in isotropic homogeneous turbulent flows [96]. However, local parametrizations are inaccurate; for example in the case of anisotropic nonhomogeneous dynamics [96,129], for correcting global error for coarse spectral discretizations [15], or in large-scale climate models [40,100].…”
Section: Related Workmentioning
confidence: 99%