New Approach to Calculation of Atmospheric Model Physics: Accurate and Fast Neural Network Emulation of Longwave Radiation in a Climate Model

Krasnopolsky, Vladimir M.; Fox‐Rabinovitz, Michael S.; Chalikov, Dmitry

doi:10.1175/mwr2923.1

Cited by 179 publications

(170 citation statements)

References 20 publications

Supporting

Mentioning

148

Contrasting

Unclassified

Order By: Relevance

“…In particular, neural networks (NN) have exploded in popularity in recent years [Goodfellow et al, 2016]. In an early paper, Krasnopolsky et al [2005] showed that a NN can emulate a realistic radiative transfer code but with much lower computational expense. They, then, trained a NN-based cumulus parameterization using a limited-area CRM, and showed that the scheme could accurately diagnose cloud fractions and precipitation [Krasnopolsky et al, 2010[Krasnopolsky et al, , 2013.…”

Section: Introductionmentioning

confidence: 99%

Prognostic validation of a neural network unified physics parameterization

Brenowitz¹,

Bretherton²

2018

Preprint

View full text Add to dashboard Cite

Key Points:• A neural network based unified physics parameterization is trained on a near-global aqua-planet simulation from a cloud-resolving model.• A numerically stable scheme is trained by minimizing the prediction error accumulated over multiple timesteps.• Prognostic single column simulations with the neural network scheme closely match the target data. * Corresponding author: Noah Brenowitz, nbren12@uw.edu -1- AbstractWeather and climate models approximate diabatic and sub-grid-scale processes in terms of grid-scale variables using parameterizations. Current parameterizations are designed by humans based on physical understanding, observations and process modeling. As a result, they are numerically efficient and interpretable, but potentially over-simplified. However, the advent of global high-resolution simulations and observations enables a more robust approach based on machine learning. In this letter, a neural network (NN) based parameterization is trained using a near-global aquaplanet simulation with a 4 km resolution (NGAqua). The NN predicts the apparent sources of heat and moisture averaged onto (160 km) 2 grid boxes. A numerically stable scheme is obtained by minimizing the prediction error over multiple timesteps rather than single one. In prognostic single column model tests, this scheme matches both the fluctuations and equilibrium of NG-Aqua simulation better than the Community Atmosphere Model does.

show abstract

Section: Introductionmentioning

confidence: 99%

Prognostic validation of a neural network unified physics parameterization

Brenowitz¹,

Bretherton²

2018

Preprint

View full text Add to dashboard Cite

show abstract

“…Because of the complexity of the physical processes involved and the complexity of their mathematical and numerical representations, some of these parameterizations are the most time-consuming components of GCMs. We have developed NN emulations for the most time-consuming part of model physics: model radiation [9][10][11][12][13]. Because, as it was mentioned above, a physically based parameterization is represented by a single mapping, we successfully used a single NN to emulate a physically based parameterization.…”

Section: Introductionmentioning

confidence: 99%

Using Ensemble of Neural Networks to Learn Stochastic Convection Parameterizations for Climate and Numerical Weather Prediction Models from Data Simulated by a Cloud Resolving Model

Krasnopolsky

Fox‐Rabinovitz

Belochitski

2013

Advances in Artificial Neural Systems

Self Cite

125

120

View full text Add to dashboard Cite

A novel approach based on the neural network (NN) ensemble technique is formulated and used for development of a NN stochastic convection parameterization for climate and numerical weather prediction (NWP) models. This fast parameterization is built based on learning from data simulated by a cloud-resolving model (CRM) initialized with and forced by the observed meteorological data available for 4-month boreal winter from November 1992 to February 1993. CRM-simulated data were averaged and processed to implicitly define a stochastic convection parameterization. This parameterization is learned from the data using an ensemble of NNs. The NN ensemble members are trained and tested. The inherent uncertainty of the stochastic convection parameterization derived following this approach is estimated. The newly developed NN convection parameterization has been tested in National Center of Atmospheric Research (NCAR) Community Atmospheric Model (CAM). It produced reasonable and promising decadal climate simulations for a large tropical Pacific region. The extent of the adaptive ability of the developed NN parameterization to the changes in the model environment is briefly discussed. This paper is devoted to a proof of concept and discusses methodology, initial results, and the major challenges of using the NN technique for developing convection parameterizations for climate and NWP models.

show abstract

“…Chevallier et al (1998) and Krasnopolsky et al (2005) have developed neural network schemes that can emulate radiative transfer calculations with much greater efficiency than correlated-k methods. These may then be used at much higher sampling resolutions.…”

Section: Introductionmentioning

confidence: 99%

Two fast radiative transfer methods to improve the temporal sampling of clouds in numerical weather prediction and climate models

Manners

Thelen

Petch

et al. 2009

Quart J Royal Meteoro Soc

View full text Add to dashboard Cite

ABSTRACT:The high computational cost of calculating the radiative heating rates in numerical weather prediction (NWP) and climate models requires that calculations are made infrequently, leading to poor sampling of the fast-changing cloud field and a poor representation of the feedback that would occur. This paper presents two related schemes for improving the temporal sampling of the cloud field. Firstly, the 'split time-stepping' scheme takes advantage of the independent nature of the monochromatic calculations of the 'correlated-k' method to split the calculation into gaseous absorption terms that are highly dependent on changes in cloud (the optically thin terms) and those that are not (optically thick). The small number of optically thin terms can then be calculated more often to capture changes in the grey absorption and scattering associated with cloud droplets and ice crystals. Secondly, the 'incremental time-stepping' scheme uses a simple radiative transfer calculation using only one or two monochromatic calculations representing the optically thin part of the atmospheric spectrum. These are found to be sufficient to represent the heating rate increments caused by changes in the cloud field, which can then be added to the last full calculation of the radiation code. We test these schemes in an operational forecast model configuration and find a significant improvement is achieved, for a small computational cost, over the current scheme employed at the Met Office. The 'incremental time-stepping' scheme is recommended for operational use, along with a new scheme to correct the surface fluxes for the change in solar zenith angle between radiation calculations.

show abstract

New Approach to Calculation of Atmospheric Model Physics: Accurate and Fast Neural Network Emulation of Longwave Radiation in a Climate Model

Cited by 179 publications

References 20 publications

Prognostic validation of a neural network unified physics parameterization

Prognostic validation of a neural network unified physics parameterization

Using Ensemble of Neural Networks to Learn Stochastic Convection Parameterizations for Climate and Numerical Weather Prediction Models from Data Simulated by a Cloud Resolving Model

Two fast radiative transfer methods to improve the temporal sampling of clouds in numerical weather prediction and climate models

Contact Info

Product

Resources

About