Tree approximation of the long wave radiation parameterization in the NCAR CAM global climate model

Belochitski, A.; Binev, Peter; DeVore, Ronald A.; Fox‐Rabinovitz, Michael S.; Krasnopolsky, Vladimir M.; Lamby, Philipp

doi:10.1016/j.cam.2011.07.013

Cited by 34 publications

(42 citation statements)

References 7 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Second, the subgrid corrections are calculated independently for each physical process rather than for all processes together as in previous studies 12,14,15 which allows for an ML parameterization structure that is motivated by physics and the calculation of the precipitation rate from the predicted tendencies. Third, we use an RF to learn from a high-resolution model whereas NNs have been used in previous studies that learned from a high-resolution model 10,12,14,15 and RFs were used only to emulate conventional parameterizations 13,34 . Parameterizations based on an RF have advantages in that their predictions automatically satisfy physical properties in the training data (without being imposed explicitly 23 ) and they make conservative predictions for samples outside of the training data which may help with the robustness of their online performance.…”

Section: Discussionmentioning

confidence: 99%

Stable machine-learning parameterization of subgrid processes for climate modeling at a range of resolutions

2020

View full text Add to dashboard Cite

Global climate models represent small-scale processes such as convection using subgrid models known as parameterizations, and these parameterizations contribute substantially to uncertainty in climate projections. Machine learning of new parameterizations from highresolution model output is a promising approach, but such parameterizations have been prone to issues of instability and climate drift, and their performance for different grid spacings has not yet been investigated. Here we use a random forest to learn a parameterization from coarse-grained output of a three-dimensional high-resolution idealized atmospheric model. The parameterization leads to stable simulations at coarse resolution that replicate the climate of the high-resolution simulation. Retraining for different coarse-graining factors shows the parameterization performs best at smaller horizontal grid spacings. Our results yield insights into parameterization performance across length scales, and they also demonstrate the potential for learning parameterizations from global high-resolution simulations that are now emerging.

show abstract

Section: Discussionmentioning

confidence: 99%

Stable machine-learning parameterization of subgrid processes for climate modeling at a range of resolutions

2020

View full text Add to dashboard Cite

show abstract

“…A promising and increasingly explored approach to accelerate or improve parametrizations is the use of machine learning techniques [7]. The application of machine learning to accelerate expensive radiative transfer computations was one of the first uses of machine learning in forward modelling in the atmospheric sciences [8], and a range of studies have used machine learning to predict vertical profiles of longwave [8][9][10][11][12][13] and shortwave [10,11,13] radiative fluxes in weather and climate models. These end-to-end approaches, i.e.…”

Section: Introductionmentioning

confidence: 99%

Predicting atmospheric optical properties for radiative transfer computations using neural networks

Veerman

Pincus

Stoffer

et al. 2021

Phil. Trans. R. Soc. A.

View full text Add to dashboard Cite

The radiative transfer equations are well known, but radiation parametrizations in atmospheric models are computationally expensive. A promising tool for accelerating parametrizations is the use of machine learning techniques. In this study, we develop a machine learning-based parametrization for the gaseous optical properties by training neural networks to emulate a modern radiation parametrization (RRTMGP). To minimize computa- tional costs, we reduce the range of atmospheric conditions for which the neural networks are applicable and use machine-specific optimized BLAS functions to accelerate matrix computations. To generate training data, we use a set of randomly perturbed atmospheric profiles and calculate optical properties using RRTMGP. Predicted optical properties are highly accurate and the resulting radiative fluxes have average errors within 0.5 W m −2 compared to RRTMGP. Our neural network-based gas optics parametrization is up to four times faster than RRTMGP, depending on the size of the neural networks. We further test the trade-off between speed and accuracy by training neural networks for the narrow range of atmospheric conditions of a single large-eddy simulation, so smaller and therefore faster networks can achieve a desired accuracy. We conclude that our machine learning-based parametrization can speed-up radiative transfer computations while retaining high accuracy. This article is part of the theme issue ‘Machine learning for weather and climate modelling’.

show abstract

“…Krasnopolsky et al (2010) presented impressive results for an emulator for the Rapid Radiative Transfer Model for General Circulation Models (RRTMG; Clough et al, 2005; Iacono et al, 2008), which improved computational speed by 16–60 times in comparison to the original scheme, while preserving long‐term (17‐yr) stability. In Belochitski et al (2011), the NN‐based radiation emulator provided better performance than the emulators based on the Classification and Regression Tree (CART). Recently, Pal et al (2019) achieved a tenfold speed improvement and 90–95% accuracy using a deep neural network (DNN), indicating a greater computational burden in DNN.…”

Section: Introductionmentioning

confidence: 99%

Evaluation of Neural Network Emulations for Radiation Parameterization in Cloud Resolving Model

Roh

Song

2020

Geophysical Research Letters

View full text Add to dashboard Cite

This study evaluated the forecast performance of neural network (NN)-based radiation emulators with 300 to 56 neurons developed under the cloud-resolving simulation. These emulators are 20-100 times faster than the original parameterization and express evolutionary features well for 6 hr. The results suggest that the frequent use of an NN emulator can improve not only computational speed but also forecasting accuracy in comparison to the infrequent use of original radiation parameterization, which is commonly used for speedup but can induce numerical instability as a result of imbalance with other processes. The forecast error of the emulator results was much improved in comparison with that for infrequent radiation runs with similar computational cost. The 56-neuron emulator results were even more accurate than the infrequent runs, which had a computational cost five times higher. The speed and accuracy advantages of radiation emulators can be utilized for weather forecasting. Plain Language Summary Radiative transfer calculations in weather and climate models often impose computational challenges because of the complexity of radiation processes. Empirical emulators based on NN have been developed to mimic radiation parameterization while reducing computational cost. The accuracy in those studies has not been strictly evaluated because the emulator cannot outpace the original radiation parameterization in terms of accuracy. However, the emulators developed in this study showed advantages both the computational cost and forecast accuracy. These advantages of radiation emulator make them useful for weather forecasting. The necessity of a trade-off between speed and accuracy in radiation calculations has resulted in the search for alternative approaches, such as a data-driven radiation emulator based on neural networks (NNs), which achieves considerable improvement in speed with reasonable accuracy. Chevallier et al. (1998, 2000) first attempted NN-based longwave (LW) radiation emulation for the European Centre for Medium-Range Weather Forecasts (ECMWF) models. The NN-based LW/shortwave (SW) emulators have been also developed for the Community Atmosphere Model (CAM), the Climate Forecast System (CFS), and the Super-Parameterized Energy Exascale Earth System Model (SP-E3SM) in various studies (Belochitski et al.

show abstract

Tree approximation of the long wave radiation parameterization in the NCAR CAM global climate model

Cited by 34 publications

References 7 publications

Stable machine-learning parameterization of subgrid processes for climate modeling at a range of resolutions

Stable machine-learning parameterization of subgrid processes for climate modeling at a range of resolutions

Predicting atmospheric optical properties for radiative transfer computations using neural networks

Evaluation of Neural Network Emulations for Radiation Parameterization in Cloud Resolving Model

Contact Info

Product

Resources

About