Peter Korn
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Peter Korn has a total of 147 co-authors. Here are the top co-authors by number of shared publications:
- Tatiana Ilyina
Max Planck Institute for Marine Microbiology, Bremen, Germany
7 shared publications - Johann Jungclaus
Max Planck Institute for Meteorology
6 shared publications - Oliver Gutjahr
Max Planck Institute for Meteorology, Hamburg, Germany
6 shared publications - Helmuth Haak
Max Planck Institute for Meteorology
5 shared publications - Nils Brüggemann
Max Planck Institute for Meteorology Hamburg Germany
5 shared publications - Stefan Hagemann
Helmholtz-Zentrum Geesthacht
5 shared publications
Latest Publications
- Open Access
- 3 August 2026
- Published by Wiley
GPU Porting of the ICON Ocean Model: Performance and Possibilities for Submesoscale Climate Simulations
Running global submesoscale-rich ocean simulations over climate time scales of order 30 years requires a throughput of at least one simulated year per day (SYPD). With the ocean component of the ICON Earth System Model running on CPU architectures, such simulations were previously unfeasible.We present the GPU port of the ICON ocean model using OpenACC directives, which yields substantial performance gains in ocean climate modeling. On NVIDIA A100 GPUs, the complete model time loop runs 5-9 times faster than on AMD EPYC 7763 CPUs using the same number of nodes, with individual subcomponents achieving speedups of 8-12 times. NVIDIA Grace-Hopper (GH200) processors provide an additional factor of 1.5-2 over the A100 GPUs. At the lower end of the scaling curve, GPU runs achieve the same throughput as CPU runs while consuming only 1/4 of the energy. A global submesoscale-rich simulation at 1.25 km resolution (R2B11, approximately 336 x 10^6 cells) achieves about 1 SYPD on 160 Grace-Hopper nodes (640 GPUs), crossing the 1-SYPD-at-1.25-km threshold required to complete a 30-year climate integration within one month of wall-clock time.Profiling reveals that the conjugate gradient solver for the free surface equation is the primary remaining bottleneck, consuming 45% of GPU time with poor arithmetic intensity and scaling behavior. This GPU port makes ICON ocean the first ocean component of an operational, CMIP-participating Earth System Model running efficiently on GPU architectures, positioning it for next-generation HPC systems and the upcoming CMIP7 climate projections.
- 3 August 2026
- Published by Wiley
Open AccessGPU Porting of the ICON Ocean Model: Performance and Possibilities for Submesoscale Climate Simulations
Running global submesoscale-rich ocean simulations over climate time scales of order 30 years requires a throughput of at least one simulated year per day (SYPD). With the ocean component of the ICON Earth System Model running on CPU architectures, such simulations were previously unfeasible.We present the GPU port of the ICON ocean model using OpenACC directives, which yields substantial performance gains in ocean climate modeling. On NVIDIA A100 GPUs, the complete model time loop runs 5-9 times faster than on AMD EPYC 7763 CPUs using the same number of nodes, with individual subcomponents achieving speedups of 8-12 times. NVIDIA Grace-Hopper (GH200) processors provide an additional factor of 1.5-2 over the A100 GPUs. At the lower end of the scaling curve, GPU runs achieve the same throughput as CPU runs while consuming only 1/4 of the energy. A global submesoscale-rich simulation at 1.25 km resolution (R2B11, approximately 336 x 10^6 cells) achieves about 1 SYPD on 160 Grace-Hopper nodes (640 GPUs), crossing the 1-SYPD-at-1.25-km threshold required to complete a 30-year climate integration within one month of wall-clock time.Profiling reveals that the conjugate gradient solver for the free surface equation is the primary remaining bottleneck, consuming 45% of GPU time with poor arithmetic intensity and scaling behavior. This GPU port makes ICON ocean the first ocean component of an operational, CMIP-participating Earth System Model running efficiently on GPU architectures, positioning it for next-generation HPC systems and the upcoming CMIP7 climate projections.
- Open Access
- 19 June 2026
- Published by American Geophysical Union (AGU) in Journal of Geophysical Research: Oceans
- Vol. 131 (6)
Diagnosing Kinetic Energy Scaling Using Lagrangian and Eulerian Metrics in Different Dynamical Regimes of the North Atlantic
Understanding the multiscale dynamics of ocean surface turbulence requires diagnostics that reliably characterize variability at scales both larger and smaller than the Rossby radius of deformation. This study evaluates the consistency and interpretability of second‐order velocity structure functions derived from Eulerian and Lagrangian frameworks using high‐resolution simulations of the North Atlantic. Synthetic surface drifters were deployed following a Sierpiński Carpet fractal pattern to achieve efficient scale‐aware Lagrangian sampling across seven dynamically distinct subdomains, while Eulerian diagnostics were computed from representative subsets of grid‐point pairs. Kinetic energy spectra and structure functions were compared across spatial scales to assess the agreement with theoretical expectations. The relationship between spectral and structure‐function slopes exhibits clear scale dependence: correlation between the two diagnostics is higher at scales larger than the Rossby radius, reflecting coherent mesoscale dynamics, whereas at smaller scales the correspondence weakens, although Lagrangian structure functions remain slightly more consistent with theoretical scaling expectations. Spectral slopes are generally steeper than structure‐function slopes, while Lagrangian slopes exceed Eulerian ones within the structure‐function framework. These results demonstrate that, when properly sampled, Lagrangian structure functions provide a robust and scalable complement to Eulerian analyses for diagnosing surface turbulence. The combined diagnostics reveal distinct turbulence regimes, with Gulf Stream regions characterized by flatter, frontal‐dominated scaling and regions outside the jet exhibiting steeper, quasi‐geostrophic behavior. Their joint use improves the interpretation of scale‐dependent energy distributions and strengthens the methodological foundation for future observational and modeling studies of upper‐ocean dynamics.
- 19 June 2026
- Published by American Geophysical Union (AGU) in Journal of Geophysical Research: Oceans
- Vol. 131 (6)
Open AccessDiagnosing Kinetic Energy Scaling Using Lagrangian and Eulerian Metrics in Different Dynamical Regimes of the North Atlantic
Understanding the multiscale dynamics of ocean surface turbulence requires diagnostics that reliably characterize variability at scales both larger and smaller than the Rossby radius of deformation. This study evaluates the consistency and interpretability of second‐order velocity structure functions derived from Eulerian and Lagrangian frameworks using high‐resolution simulations of the North Atlantic. Synthetic surface drifters were deployed following a Sierpiński Carpet fractal pattern to achieve efficient scale‐aware Lagrangian sampling across seven dynamically distinct subdomains, while Eulerian diagnostics were computed from representative subsets of grid‐point pairs. Kinetic energy spectra and structure functions were compared across spatial scales to assess the agreement with theoretical expectations. The relationship between spectral and structure‐function slopes exhibits clear scale dependence: correlation between the two diagnostics is higher at scales larger than the Rossby radius, reflecting coherent mesoscale dynamics, whereas at smaller scales the correspondence weakens, although Lagrangian structure functions remain slightly more consistent with theoretical scaling expectations. Spectral slopes are generally steeper than structure‐function slopes, while Lagrangian slopes exceed Eulerian ones within the structure‐function framework. These results demonstrate that, when properly sampled, Lagrangian structure functions provide a robust and scalable complement to Eulerian analyses for diagnosing surface turbulence. The combined diagnostics reveal distinct turbulence regimes, with Gulf Stream regions characterized by flatter, frontal‐dominated scaling and regions outside the jet exhibiting steeper, quasi‐geostrophic behavior. Their joint use improves the interpretation of scale‐dependent energy distributions and strengthens the methodological foundation for future observational and modeling studies of upper‐ocean dynamics.
- Open Access
- 10 November 2025
- Published by European Mathematical Society - EMS - Publishing House GmbH in Oberwolfach Reports
- Vol. 22 (2) , 1215-1270
Mathematical Advances in Geophysical Fluid Dynamics
The workshop “Mathematical Advances in Geophysical Fluid Dynamics” addressed recent advances in modeling, analytical, computational and stochastical studies of geophysical models. In particular addressed are atmosphere-, ocean- and sea-ice models, well-posedness and analysis of relevant geophysical fluid regimes and models, including stochastically forced equations, boundary layers as well as coupled models.
- 10 November 2025
- Published by European Mathematical Society - EMS - Publishing House GmbH in Oberwolfach Reports
- Vol. 22 (2) , 1215-1270
Open AccessMathematical Advances in Geophysical Fluid Dynamics
The workshop “Mathematical Advances in Geophysical Fluid Dynamics” addressed recent advances in modeling, analytical, computational and stochastical studies of geophysical models. In particular addressed are atmosphere-, ocean- and sea-ice models, well-posedness and analysis of relevant geophysical fluid regimes and models, including stochastically forced equations, boundary layers as well as coupled models.
- Open Access
- 5 November 2025
- Published by Wiley
Diagnosing Kinetic Energy Scaling using Lagrangian and Eulerian Metrics in Different Dynamical Regimes of the North Atlantic
Understanding the multiscale dynamics of ocean surface turbulence requires diagnostics that can reliably characterize variability at scales both larger and smaller than the Rossby radius of deformation. This study evaluates the consistency and interpretability of second-order velocity structure functions derived from Eulerian and Lagrangian frameworks using high-resolution numerical simulations of the North Atlantic. Synthetic surface drifters were deployed following a Sierpiński Carpet fractal pattern to achieve efficient and scale-aware Lagrangian sampling across seven dynamically distinct subdomains, while Eulerian diagnostics were computed from statistically representative subsets of grid-point pairs. Kinetic energy spectra and structure functions were compared across spatial scales to assess their agreement with theoretical expectations. The relationship between structure function and spectral slopes is generally stronger at scales smaller than the Rossby radius, where Lagrangian structure functions show slightly better correspondence. The linear correlation between spectral and structure function slopes is higher at larger scales, suggesting that coherent, slowly evolving flow features are well captured by both approaches. Spectral slopes tend to be steeper than those from structure functions, while Lagrangian slopes are generally steeper than Eulerian ones within the structure function framework. These results demonstrate that, when properly sampled, Lagrangian structure functions provide a robust and scalable complement to traditional Eulerian analyses for diagnosing surface turbulence. The combined use of both frameworks improves interpretation of scale-dependent energy distributions and strengthens the methodological foundation for future observational and modeling studies of upper-ocean dynamics.
- 5 November 2025
- Published by Wiley
Open AccessDiagnosing Kinetic Energy Scaling using Lagrangian and Eulerian Metrics in Different Dynamical Regimes of the North Atlantic
Understanding the multiscale dynamics of ocean surface turbulence requires diagnostics that can reliably characterize variability at scales both larger and smaller than the Rossby radius of deformation. This study evaluates the consistency and interpretability of second-order velocity structure functions derived from Eulerian and Lagrangian frameworks using high-resolution numerical simulations of the North Atlantic. Synthetic surface drifters were deployed following a Sierpiński Carpet fractal pattern to achieve efficient and scale-aware Lagrangian sampling across seven dynamically distinct subdomains, while Eulerian diagnostics were computed from statistically representative subsets of grid-point pairs. Kinetic energy spectra and structure functions were compared across spatial scales to assess their agreement with theoretical expectations. The relationship between structure function and spectral slopes is generally stronger at scales smaller than the Rossby radius, where Lagrangian structure functions show slightly better correspondence. The linear correlation between spectral and structure function slopes is higher at larger scales, suggesting that coherent, slowly evolving flow features are well captured by both approaches. Spectral slopes tend to be steeper than those from structure functions, while Lagrangian slopes are generally steeper than Eulerian ones within the structure function framework. These results demonstrate that, when properly sampled, Lagrangian structure functions provide a robust and scalable complement to traditional Eulerian analyses for diagnosing surface turbulence. The combined use of both frameworks improves interpretation of scale-dependent energy distributions and strengthens the methodological foundation for future observational and modeling studies of upper-ocean dynamics.
- Open Access
- 19 December 2024
- Published by IOP Publishing in Machine Learning: Science and Technology
- Vol. 6 (1)
Dynamic Deep Learning Based Super-Resolution For The Shallow Water Equations
Correctly capturing the transition to turbulence in a barotropic instability requires fine spatial resolution. To reduce computational cost, we propose a dynamic super-resolution approach where a transient simulation on a coarse mesh is frequently corrected using a U-net-type neural network. For the nonlinear shallow water equations, we demonstrate that a simulation with the ICON-O ocean model with a 20 km resolution plus dynamic super-resolution trained on a 2.5 km resolution achieves discretization errors comparable to a simulation with 10 km resolution. The neural network, originally developed for image-based super-resolution in post-processing, is trained to compute the difference between solutions on both meshes and is used to correct the coarse mesh solution every 12 h. We show that the ML-corrected coarse solution correctly maintains a balanced flow and captures the transition to turbulence in line with the higher resolution simulation. After a 8 d simulation, the L2-error of the corrected run is similar to a simulation run on the finer mesh. While mass is conserved in the corrected runs, we observe some spurious generation of kinetic energy.
- 19 December 2024
- Published by IOP Publishing in Machine Learning: Science and Technology
- Vol. 6 (1)
Open AccessDynamic Deep Learning Based Super-Resolution For The Shallow Water Equations
Correctly capturing the transition to turbulence in a barotropic instability requires fine spatial resolution. To reduce computational cost, we propose a dynamic super-resolution approach where a transient simulation on a coarse mesh is frequently corrected using a U-net-type neural network. For the nonlinear shallow water equations, we demonstrate that a simulation with the ICON-O ocean model with a 20 km resolution plus dynamic super-resolution trained on a 2.5 km resolution achieves discretization errors comparable to a simulation with 10 km resolution. The neural network, originally developed for image-based super-resolution in post-processing, is trained to compute the difference between solutions on both meshes and is used to correct the coarse mesh solution every 12 h. We show that the ML-corrected coarse solution correctly maintains a balanced flow and captures the transition to turbulence in line with the higher resolution simulation. After a 8 d simulation, the L2-error of the corrected run is similar to a simulation run on the finer mesh. While mass is conserved in the corrected runs, we observe some spurious generation of kinetic energy.
- 16 December 2024
- Published by Society for Industrial & Applied Mathematics (SIAM) in SIAM Journal on Mathematical Analysis
- Vol. 56 (6) , 8011-8036
Global Well-Posedness of the Primitive Equations of Large-Scale Ocean Dynamics with the Gent–McWilliams–Redi Eddy Parametrization Model
We prove global well-posedness of the ocean primitive equations coupled to advection-diffusion equations of the oceanic tracers temperature and salinity that are supplemented by the eddy parametrization model due to Gent, McWilliams, and Redi. This parametrization forms a milestone in global ocean modeling and constitutes a central part of any general ocean circulation model computation. The eddy parametrization adds a secondary transport velocity to the tracer equation and renders the original Laplacian operators in the advection-diffusion equations nonlinear, with a diffusion matrix that depends via the equation of state in a nonlinear fashion on both tracers simultaneously. The eddy parametrization of Gent, McWilliams, and Redi augments the complexity of the mathematical analysis of the whole system which we present here. We show first that weak solutions exist globally in time, provided the parametrization uses a regularized density. Then we prove by a detailed analysis of the eddy operators the global well-posedness. Our results apply also to the “small-slope approximation” that is commonly used in global ocean simulations.
- 16 December 2024
- Published by Society for Industrial & Applied Mathematics (SIAM) in SIAM Journal on Mathematical Analysis
- Vol. 56 (6) , 8011-8036
Global Well-Posedness of the Primitive Equations of Large-Scale Ocean Dynamics with the Gent–McWilliams–Redi Eddy Parametrization Model
We prove global well-posedness of the ocean primitive equations coupled to advection-diffusion equations of the oceanic tracers temperature and salinity that are supplemented by the eddy parametrization model due to Gent, McWilliams, and Redi. This parametrization forms a milestone in global ocean modeling and constitutes a central part of any general ocean circulation model computation. The eddy parametrization adds a secondary transport velocity to the tracer equation and renders the original Laplacian operators in the advection-diffusion equations nonlinear, with a diffusion matrix that depends via the equation of state in a nonlinear fashion on both tracers simultaneously. The eddy parametrization of Gent, McWilliams, and Redi augments the complexity of the mathematical analysis of the whole system which we present here. We show first that weak solutions exist globally in time, provided the parametrization uses a regularized density. Then we prove by a detailed analysis of the eddy operators the global well-posedness. Our results apply also to the “small-slope approximation” that is commonly used in global ocean simulations.
- Top Cited PapersOpen Access
- 7 November 2024
- Published by American Speech Language Hearing Association in Journal of Speech, Language, and Hearing Research
- Vol. 67 (11) , 4162-4175
Community-Supported Shared Infrastructure in Support of Speech Accessibility
Purpose: The Speech Accessibility Project (SAP) intends to facilitate research and development in automatic speech recognition (ASR) and other machine learning tasks for people with speech disabilities. The purpose of this article is to introduce this project as a resource for researchers, including baseline analysis of the first released data package. Method: The project aims to facilitate ASR research by collecting, curating, and distributing transcribed U.S. English speech from people with speech and/or language disabilities. Participants record speech from their place of residence by connecting their personal computer, cell phone, and assistive devices, if needed, to the SAP web portal. All samples are manually transcribed, and 30 per participant are annotated using differential diagnostic pattern dimensions. For purposes of ASR experiments, the participants have been randomly assigned to a training set, a development set for controlled testing of a trained ASR, and a test set to evaluate ASR error rate. Results: The SAP 2023-10-05 Data Package contains the speech of 211 people with dysarthria as a correlate of Parkinson's disease, and the associated test set contains 42 additional speakers. A baseline ASR, with a word error rate of 3.4% for typical speakers, transcribes test speech with a word error rate of 36.3%. Fine-tuning reduces the word error rate to 23.7%. Conclusions: Preliminary findings suggest that a large corpus of dysarthric and dysphonic speech has the potential to significantly improve speech technology for people with disabilities. By providing these data to researchers, the SAP intends to significantly accelerate research into accessible speech technology. Supplemental Material: https://doi.org/10.23641/asha.27078079
- 7 November 2024
- Published by American Speech Language Hearing Association in Journal of Speech, Language, and Hearing Research
- Vol. 67 (11) , 4162-4175
Top Cited PapersOpen AccessCommunity-Supported Shared Infrastructure in Support of Speech Accessibility
Purpose: The Speech Accessibility Project (SAP) intends to facilitate research and development in automatic speech recognition (ASR) and other machine learning tasks for people with speech disabilities. The purpose of this article is to introduce this project as a resource for researchers, including baseline analysis of the first released data package. Method: The project aims to facilitate ASR research by collecting, curating, and distributing transcribed U.S. English speech from people with speech and/or language disabilities. Participants record speech from their place of residence by connecting their personal computer, cell phone, and assistive devices, if needed, to the SAP web portal. All samples are manually transcribed, and 30 per participant are annotated using differential diagnostic pattern dimensions. For purposes of ASR experiments, the participants have been randomly assigned to a training set, a development set for controlled testing of a trained ASR, and a test set to evaluate ASR error rate. Results: The SAP 2023-10-05 Data Package contains the speech of 211 people with dysarthria as a correlate of Parkinson's disease, and the associated test set contains 42 additional speakers. A baseline ASR, with a word error rate of 3.4% for typical speakers, transcribes test speech with a word error rate of 36.3%. Fine-tuning reduces the word error rate to 23.7%. Conclusions: Preliminary findings suggest that a large corpus of dysarthric and dysphonic speech has the potential to significantly improve speech technology for people with disabilities. By providing these data to researchers, the SAP intends to significantly accelerate research into accessible speech technology. Supplemental Material: https://doi.org/10.23641/asha.27078079
- Open Access
- 21 June 2024
- Published by American Geophysical Union (AGU) in Journal of Advances in Modeling Earth Systems
- Vol. 16 (6)
Parameterized Internal Wave Mixing in Three Ocean General Circulation Models
The non‐local model of mixing based on internal wave breaking, IDEMIX, is implemented as an enhancement of a turbulent kinetic energy closure model in three non‐eddy resolving general circulation ocean models that differ in the discretization and choice of computational grids. In IDEMIX internal wave energy is generated by an energy flux resulting from near‐inertial waves induced by wind forcing at the surface, and at the bottom, by an energy flux that parameterizes the transfer of energy between baroclinic and barotropic tides. In all model simulations with IDEMIX, the mixing work is increased compared to the reference solutions without IDEMIX, reaching values in better agreement with finestructure observations. Furthermore, the horizontal structure of the mixing work is more realistic as a consequence of the heterogeneous forcing functions. All models with IDEMIX simulate deeper thermocline depths related to stronger shallow overturning cells in the Indo‐Pacific. In the North Atlantic, deeper mixed layers in simulations with IDEMIX are associated with an increased Atlantic overturning circulation and an increase of northward heat transports toward more realistic values. The response of the deep Indo‐Pacific overturning circulation and the weak bottom cell of the Atlantic to the inclusion of IDEMIX is incoherent between the models, suggesting that additional unidentified processes and numerical mixing may confound the analysis. Applying different tidal forcing functions leads to simulation differences that are small compared to differences between the different models or between simulations with IDEMIX and without IDEMIX.
- 21 June 2024
- Published by American Geophysical Union (AGU) in Journal of Advances in Modeling Earth Systems
- Vol. 16 (6)
Open AccessParameterized Internal Wave Mixing in Three Ocean General Circulation Models
The non‐local model of mixing based on internal wave breaking, IDEMIX, is implemented as an enhancement of a turbulent kinetic energy closure model in three non‐eddy resolving general circulation ocean models that differ in the discretization and choice of computational grids. In IDEMIX internal wave energy is generated by an energy flux resulting from near‐inertial waves induced by wind forcing at the surface, and at the bottom, by an energy flux that parameterizes the transfer of energy between baroclinic and barotropic tides. In all model simulations with IDEMIX, the mixing work is increased compared to the reference solutions without IDEMIX, reaching values in better agreement with finestructure observations. Furthermore, the horizontal structure of the mixing work is more realistic as a consequence of the heterogeneous forcing functions. All models with IDEMIX simulate deeper thermocline depths related to stronger shallow overturning cells in the Indo‐Pacific. In the North Atlantic, deeper mixed layers in simulations with IDEMIX are associated with an increased Atlantic overturning circulation and an increase of northward heat transports toward more realistic values. The response of the deep Indo‐Pacific overturning circulation and the weak bottom cell of the Atlantic to the inclusion of IDEMIX is incoherent between the models, suggesting that additional unidentified processes and numerical mixing may confound the analysis. Applying different tidal forcing functions leads to simulation differences that are small compared to differences between the different models or between simulations with IDEMIX and without IDEMIX.
- 14 December 2023
- Published by Springer Nature
- p. 113-130
Numerical Simulation of an idealized coupled Ocean-Atmosphere Climate Model
- 14 December 2023
- Published by Springer Nature
- p. 113-130
Numerical Simulation of an idealized coupled Ocean-Atmosphere Climate Model
- Open Access
- 3 August 2023
- Published by IOP Publishing in Nonlinearity
- Vol. 36 (9) , 4862-4903
An implementation of Hasselmann’s paradigm for stochastic climate modelling based on stochastic Lie transport *
A generic approach to stochastic climate modelling is developed for the example of an idealised Atmosphere-Ocean model that rests upon Hasselmann’s paradigm for stochastic climate models. Namely, stochasticity is incorporated into the fast moving atmospheric component of an idealised coupled model by means of stochastic Lie transport, while the slow moving ocean model remains deterministic. More specifically the stochastic model stochastic advection by Lie transport (SALT) is constructed by introducing stochastic transport into the material loop in Kelvin’s circulation theorem. The resulting stochastic model preserves circulation, as does the underlying deterministic climate model. A variant of SALT called Lagrangian-averaged (LA)-SALT is introduced in this paper. In LA-SALT, we replace the drift velocity of the stochastic vector field by its expected value. The remarkable property of LA-SALT is that the evolution of its higher moments are governed by deterministic equations. Our modelling approach is substantiated by establishing local existence results, first, for the deterministic climate model that couples compressible atmospheric equations to incompressible ocean equation, and second, for the two stochastic SALT and LA-SALT models.
- 3 August 2023
- Published by IOP Publishing in Nonlinearity
- Vol. 36 (9) , 4862-4903
Open AccessAn implementation of Hasselmann’s paradigm for stochastic climate modelling based on stochastic Lie transport *
A generic approach to stochastic climate modelling is developed for the example of an idealised Atmosphere-Ocean model that rests upon Hasselmann’s paradigm for stochastic climate models. Namely, stochasticity is incorporated into the fast moving atmospheric component of an idealised coupled model by means of stochastic Lie transport, while the slow moving ocean model remains deterministic. More specifically the stochastic model stochastic advection by Lie transport (SALT) is constructed by introducing stochastic transport into the material loop in Kelvin’s circulation theorem. The resulting stochastic model preserves circulation, as does the underlying deterministic climate model. A variant of SALT called Lagrangian-averaged (LA)-SALT is introduced in this paper. In LA-SALT, we replace the drift velocity of the stochastic vector field by its expected value. The remarkable property of LA-SALT is that the evolution of its higher moments are governed by deterministic equations. Our modelling approach is substantiated by establishing local existence results, first, for the deterministic climate model that couples compressible atmospheric equations to incompressible ocean equation, and second, for the two stochastic SALT and LA-SALT models.