Earth System Modeling 2.0: A Blueprint for Models That Learn From Observations and Targeted High‐Resolution Simulations

Schneider, Tapio; Lan, Shiwei; Stuart, Andrew M.; Teixeira, J.

doi:10.1002/2017gl076101

Cited by 316 publications

(327 citation statements)

References 198 publications

(264 reference statements)

Supporting

Mentioning

326

Contrasting

Order By: Relevance

“…By contrast, Rasp et al () and Brenowitz and Bretherton () use machine learning to constrain the convective tendency to high‐resolution simulation data based on a multiscale modeling framework and cloud resolving modeling, respectively. See also the work of Gentine et al () and Schneider et al ().…”

Section: Introductionmentioning

confidence: 99%

Using Radar Data to Calibrate a Stochastic Parametrization of Organized Convection

Cardoso-Bihlo

Khouider

Schumacher

et al. 2019

J Adv Model Earth Syst

View full text Add to dashboard Cite

Stochastic parameterizations are increasingly becoming skillful in representing unresolved atmospheric processes for global climate models. The stochastic multicloud model, used to simulate the life cycle of the three most common cloud types (cumulus congestus, deep convective, and stratiform) in tropical convective systems, is one example. In this model, these clouds interact with each other and with their environment according to intuitive‐probabilistic rules determined by a set of predictors, depending on the large‐scale atmospheric state and a set of transition time scale parameters. Here we use a Bayesian statistical method to infer these parameters from radar data. The Bayesian approach is applied to precipitation data collected by the Shared Mobile Atmospheric Research and Teaching Radar truck‐mounted C‐band radar located in the Maldives archipelago, while the corresponding large‐scale predictors were derived from meteorological soundings taken during the Dynamics of the Madden‐Julian Oscillation field campaign. The transition time scales were inferred from three different phases of the Madden‐Julian Oscillation (suppressed, initiation, and active) and compared with previous studies. The performance of the stochastic multicloud model is also assessed, in a stand‐alone mode, where the cloud model is forced directly by the observed predictors without feedback into the environmental variables. The results showed a wide spread in the inferred parameter values due in part to the lack of the desired sensitivity of the model to the predictors and the shortness of the training periods that did not include both active and suppressed convection phases simultaneously. Nonetheless, the resemblance of the stand‐alone simulated cloud fraction time series to the radar data is encouraging.

show abstract

Section: Introductionmentioning

confidence: 99%

Using Radar Data to Calibrate a Stochastic Parametrization of Organized Convection

Cardoso-Bihlo

Khouider

Schumacher

et al. 2019

J Adv Model Earth Syst

View full text Add to dashboard Cite

show abstract

“…Some recent work has proposed using ensemble Kalman filters [Schneider et al, 2017] and genetic algorithms [Langenbrunner and Neelin, 2017] to automatically discover the parameters in a traditional parameterization. However, parameters in traditional schemes are designed to be human-interpretable rather than machine-tunable.…”

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

“…Our results also suggest that running ultra-high-resolution ESMs utilizing detailed parameterizations (e.g., microphysics, boundary layer, and land surface) with future advances in computational resources could improve simulated mean and extreme climates (e.g., Schneider et al, 2017).…”

Section: Discussionmentioning

confidence: 68%

High‐Resolution Climate Projections for the Northeastern United States Using Dynamical Downscaling at Convection‐Permitting Scales

Kömürcü

Emanuel

Huber

et al. 2018

Earth and Space Science

View full text Add to dashboard Cite

To paraphrase former Speaker of the House Tip O'Neill, “All climate change is local”—that is, society reacts most immediately to changes in local weather such as regional heat waves and heavy rainstorms. Such phenomena are not well resolved by the current generation of coupled climate models. Here it is shown that dynamical downscaling of climate reanalyses using a high‐resolution regional model can reproduce both the means and extremes of temperature and precipitation as observed in the well‐measured northeastern United States. Given this result, the downscaling is applied to climate projections for the middle and end of the 21st century under Representative Concentration Pathway (RCP) 8.5 as well as for the historical time period to help assess regional climate impacts in the northeastern United States. The resulting high‐resolution projections are intended to support regional sustainability studies for the northeastern United States and are made publicly available.

show abstract

Earth System Modeling 2.0: A Blueprint for Models That Learn From Observations and Targeted High‐Resolution Simulations

Cited by 316 publications

References 198 publications

Using Radar Data to Calibrate a Stochastic Parametrization of Organized Convection

Using Radar Data to Calibrate a Stochastic Parametrization of Organized Convection

Prognostic validation of a neural network unified physics parameterization

High‐Resolution Climate Projections for the Northeastern United States Using Dynamical Downscaling at Convection‐Permitting Scales

Contact Info

Product

Resources

About