Neural general circulation models for weather and climate

Nature, Journal Year: 2024, Volume and Issue: 632(8027), P. 1060 - 1066

Published: July 22, 2024

General circulation models (GCMs) are the foundation of weather and climate prediction

Language: Английский

---

Exacerbated summer European warming not captured by climate models neglecting long-term aerosol changes

Communications Earth & Environment, Journal Year: 2024, Volume and Issue: 5(1)

Published: April 6, 2024

Abstract In much of western-central Europe, summer temperatures have surged three times faster than the global mean warming since 1980, yet this is not captured by most climate model simulations. Here we disentangle into thermodynamic and circulation-induced contributions, show that latter main reason why numerically simulated weaker observed. Crucially, regional models from Coordinated Regional Downscaling Experiment with constant aerosol forcings systematically strongest discrepancies observations: in these simulations, brightening associated due to reductions represented. We estimate an effect ~0.5 °C over Europe for our ensemble, discrepancy evolving aerosols increases future projections. To better reap benefits high-resolution it thus imperative represent relevant external responses across entire chain.

Language: Английский

---

Toward machine-assisted tuning avoiding the underestimation of uncertainty in climate change projections

Science Advances, Journal Year: 2023, Volume and Issue: 9(29)

Published: July 19, 2023

Documenting the uncertainty of climate change projections is a fundamental objective inter-comparison exercises organized to feed into Intergovernmental Panel on Climate Change (IPCC) reports. Usually, each modeling center contributes these with one or two configurations its model, corresponding particular choice "free parameter" values, resulting from long and often tedious "model tuning" phase. How much omitted by this selection how might readers IPCC reports users be misled omission? We show here recent machine learning approaches can transform way model tuning approached, opening simultaneous acceleration improvement parametric quantification. an automatic defined different values free parameters produce "warming worlds," all consistent present-day observations system.

Language: Английский

---

Opinion: Optimizing climate models with process knowledge, resolution, and artificial intelligence

Atmospheric chemistry and physics, Journal Year: 2024, Volume and Issue: 24(12), P. 7041 - 7062

Published: June 19, 2024

Abstract. Accelerated progress in climate modeling is urgently needed for proactive and effective change adaptation. The central challenge lies accurately representing processes that are small scale yet climatically important, such as turbulence cloud formation. These will not be explicitly resolvable the foreseeable future, necessitating use of parameterizations. We propose a balanced approach leverages strengths traditional process-based parameterizations contemporary artificial intelligence (AI)-based methods to model subgrid-scale processes. This strategy employs AI derive data-driven closure functions from both observational simulated data, integrated within encode system knowledge conservation laws. In addition, increasing resolution resolve larger fraction small-scale can aid toward improved interpretable predictions outside observed distribution. However, currently feasible horizontal resolutions limited O(10 km) because higher would impede creation ensembles calibration uncertainty quantification, sampling atmospheric oceanic internal variability, broadly exploring quantifying risks. By synergizing decades scientific development with advanced techniques, our aims significantly boost accuracy, interpretability, trustworthiness predictions.

Language: Английский

---

Explainable Offline‐Online Training of Neural Networks for Parameterizations: A 1D Gravity Wave‐QBO Testbed in the Small‐Data Regime

Geophysical Research Letters, Journal Year: 2024, Volume and Issue: 51(2)

Published: Jan. 27, 2024

Abstract There are different strategies for training neural networks (NNs) as subgrid‐scale parameterizations. Here, we use a 1D model of the quasi‐biennial oscillation (QBO) and gravity wave (GW) parameterizations testbeds. A 12‐layer convolutional NN that predicts GW forcings given wind profiles, when trained offline in big ‐ data regime (100‐year), produces realistic QBOs once coupled to model. In contrast, this small (18‐month) yields unrealistic QBOs. However, online re‐training just two layers using ensemble Kalman inversion only time‐averaged QBO statistics leads yield Fourier analysis these three NNs' kernels suggests why/how works reveals NNs primarily learn low‐pass, high‐pass, combination band‐pass filters, potentially related local non‐local dynamics propagation dissipation. These findings/strategies generally apply data‐driven other climate processes.

Language: Английский

---

Obtaining refined Euro-Mediterranean rainfall projections through regional assessment of CMIP6 General Circulation Models

Global and Planetary Change, Journal Year: 2025, Volume and Issue: unknown, P. 104725 - 104725

Published: Jan. 1, 2025

Language: Английский

---

Using Machine Learning to Generate a GISS ModelE Calibrated Physics Ensemble (CPE)

Journal of Advances in Modeling Earth Systems, Journal Year: 2025, Volume and Issue: 17(4)

Published: April 1, 2025

Abstract A neural network (NN) surrogate of the NASA GISS ModelE atmosphere (version E3) is trained on a perturbed parameter ensemble (PPE) spanning 45 physics parameters and 36 outputs. The NN leveraged in Markov Chain Monte Carlo (MCMC) Bayesian inference framework to generate second posterior constrained coined “calibrated ensemble,” or CPE. CPE members are characterized by diverse combinations are, definition, close top‐of‐atmosphere radiative balance, must broadly agree with numerous hydrologic, energy cycle forcing metrics simultaneously. Global observations cloud, environment, radiation properties (provided global satellite products) crucial for generation. explicitly accounts discrepancies (or biases) products during We demonstrate that product strongly impact calibration important model settings (e.g., convective plume entrainment rates; fall speed cloud ice). Structural improvements new E3 retained across stratocumulus simulation). Notably, improved simulation shallow cumulus Amazon rainfall while not degrading fields, an upgrade neither default nor Latin Hypercube searching achieved. Analyses initial PPE suggested several were unimportant output variation. However, many “unimportant” needed generation, result brings forefront how importance should be determined PPEs. From CPE, two 45‐dimensional configurations radiatively‐balanced, auto‐tuned atmospheres used submissions CMIP6.

Language: Английский

---

Parameterizing Vertical Mixing Coefficients in the Ocean Surface Boundary Layer Using Neural Networks

Journal of Advances in Modeling Earth Systems, Journal Year: 2023, Volume and Issue: 15(10)

Published: Oct. 1, 2023

Abstract Vertical mixing parameterizations in ocean models are formulated on the basis of physical principles that govern turbulent mixing. However, many include ad hoc components not well constrained by theory or data. One such component is eddy diffusivity model, where vertical fluxes a quantity parameterized from variable diffusion coefficient and mean gradient quantity. In this work, we improve parameterization surface boundary layer enhancing its model using data‐driven methods, specifically neural networks. The networks designed to take extrinsic intrinsic forcing parameters as input predict profile trained output data second moment closure scheme. modified scheme predicts through online inference maintains conservation standard equations, which particularly important for targeted use climate simulations. We describe development stable implementation an general circulation demonstrate enhanced outperforms predecessor reducing biases mixed‐layer depth upper stratification. Our results potential physics‐aware global models.

Language: Английский

---

Simulations of idealised 3D atmospheric flows on terrestrial planets using LFRic-Atmosphere

Geoscientific model development, Journal Year: 2023, Volume and Issue: 16(19), P. 5601 - 5626

Published: Oct. 10, 2023

Abstract. We demonstrate that LFRic-Atmosphere, a model built using the Met Office's GungHo dynamical core, is able to reproduce idealised large-scale atmospheric circulation patterns specified by several widely used benchmark recipes. This motivated rapid rate of exoplanet discovery and ever-growing need for numerical modelling characterisation their atmospheres. Here we present LFRic-Atmosphere's results tests imitating regimes commonly in community. The benchmarks include three analytic forcing cases: standard Held–Suarez test, Menou–Rauscher Earth-like Merlis–Schneider tidally locked Earth test. Qualitatively, LFRic-Atmosphere agrees well with other models shows excellent conservation properties terms total mass, angular momentum, kinetic energy. then use more realistic representation physical processes (radiation, subgrid-scale mixing, convection, clouds) configuring it four TRAPPIST-1 Habitable Atmosphere Intercomparison (THAI) scenarios. first application possible climate confirmed terrestrial exoplanet. reproduces THAI scenarios within spread existing across range key climatic variables. Our work performs seven atmospheres, justifying its future studies.

Language: Английский

---

A Perspective on the Future of CMIP

AGU Advances, Journal Year: 2024, Volume and Issue: 5(1)

Published: Feb. 1, 2024

Abstract The Coupled Model Intercomparison Project (CMIP) has demonstrated the importance of climate modeling for research and its usefulness services. latter increased CMIP's operational burden, so much that serving IPCC become animating force. Attempting to satisfy an mandate through a coordinated project diminishes both service research. Regaining initiative will require CMIP transition quasi‐operational system it developed setting. Doing would allow focus on developing international scientific agenda encourage exploit advances in modeling.

Language: Английский

---