NLNora Loose

Nora Loose

*This is an algorithmically generated author record based on publicly available information.

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article


Nora Loose has a total of 130 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Open Access

    An online-learned eddy parameterization in a differentiable shallow water model

    by Sarah Williamson, Nora Loose, William S Moses, Milan Klöewer, Valentin Churavy, Michel Schanen, Sri Hari Krishna Narayanan, Patrick Heimbach

    Ocean general circulation models rely on parameterizations to approximate subgrid-scale processes. While parameterizations are traditionally physics-based, machine learning has recently provided an alternative, data-driven approach for developing parameterizations. However, most machine learning-based parameterizations are learned offline due to the lack of differentiable general circulation models. Here, we train an eddy momentum parameterization represented by a neural network online in an idealized barotropic gyre model. We develop the model as fully differentiable in Julia with the automatic differentiation tool Enzyme. The network weights are trained using a loss function that has the same form as in adjoint-based data assimilation, with synthetic observations generated by a fine-resolution simulation. A range of diagnostics evaluate the performance of the parameterized coarse-resolution model. We find improved energetics, physically realistic dynamics, and numerical stability for multi-decadal roll-outs. Importantly, the online-learned parameterization absorbs artifacts from the numerical discretization, such as the time-stepping scheme used, which is invisible to offline-learned parameterizations. We suggest that this is key to improved numerical stability of machine-learned parameterizations. Our research suggests that online-trained neural network parameterizations can be an effective solution to the subgrid-scale parameterization and model calibration problem.

  • Open Access

    Linearized Carbonate Chemistry for Efficient Modeling of Ocean-Based Carbon Dioxide Removal

    by Nora Loose, Matthew Long, Scott Bachmann, Pierre Damien, Scott Eilerman, Alicia Karspeck, Jeroen Molemaker, Dafydd Stephenson, Michael Tyka, Abigale Wyatt

    Ocean-based carbon dioxide removal (CDR) relies on ocean models for measurement, reporting, and verification (MRV), because induced CO₂ uptake spans basin scales and decades. Explicit calculations with coupled physics–biogeochemistry models are computationally expensive, because biogeochemical models carry many tracers and each CDR intervention must be simulated separately. We introduce a CDR tracer method that represents each direct ocean removal (DOR) intervention by one passive tracer and each ocean alkalinity enhancement (OAE) intervention by two. Air–sea gas exchange is linearized using precomputed carbonate sensitivities, so that many interventions can be simulated in a single integration without a biogeochemical model. We evaluate the method against 24 explicit calculations with a 12 km ROMS configuration coupled to the Marine Biogeochemistry Library (MARBL), and reproduce all 1,380 explicit 1° CESM/MARBL calculations of existing DOR and OAE efficiency atlases in two CESM integrations. The CDR tracer method is 13.5–28× faster than explicit calculations. Its relative errors in CDR efficiency, predominantly positive owing to the linearization, peak shortly after intervention and then decay to less than 1% within a few years. In ROMS, peak errors reach 2–8% for extreme interventions but stay below 1% for plausible ones; in CESM, they average 3.5 ± 2%. Larger local errors, including in pH estimates, arise only where extreme interventions accumulate in semi-enclosed passages. The computational savings enable large ensembles across deployment scenarios and ocean models. Combined with the option to use carbonate sensitivities from different biogeochemical models or observation-based products, these ensembles support uncertainty quantification for MRV.

  • Open Access

    ROMS-Tools: Reproducible Preprocessing and Analysis for Regional Ocean Modeling with ROMS

    by Nora Loose, Tom Nicholas, Sam Maticka, Scott Eilerman, Christopher McBride, Dafydd Stephenson, Ulla Heede, Benjamin Saenz, Kristen M. Thyng, Scott Bachman, et al.

  • Open Access

    DJ4Earth: Differentiable, and Performance‐Portable Earth System Modeling via Program Transformations

    by William S. Moses, Gong Cheng, Valentin Churavy, Maximilian Gelbrecht, Milan Klöwer, Joseph Kump, Mathieu Morlighem, Sarah Williamson, Dhruv Apte, Paul Berg, et al.

    Differentiable Earth system models (ESMs) enable powerful applications such as sensitivity analysis, gradient‐based calibration, state estimation, boundary flux inversions, uncertainty quantification, and online machine learning. Reverse‐mode automatic differentiation (AD) efficiently provides gradients for such tasks, yet models have rarely included this capability because of complex, bespoke numerical algorithms. As part of the Differentiable programming in Julia for Earth system modeling (DJ4Earth) initiative, we present improved capabilities of the AD tool Enzyme.jl and the new compiler transpilation tool Reactant.jl, augmented by sophisticated checkpointing algorithms, which, together make general‐purpose AD tractable and efficient for full‐fledged ESM components written in Julia. Operating at the low‐level virtual machine intermediate representation or multi‐level intermediate representation compiler levels, these frameworks support mutable memory, custom kernels, and compiler optimizations before and after differentiation. Julia‐specific challenges related to just‐in‐time compilation and garbage collection are handled efficiently. Reactant further enables automatic performance portability across central processing units, graphics processing units, and tensor processing units, facilitating use of emerging AI‐customized high‐performance computing architectures. We demonstrate these frameworks on four Julia‐based ESM components featuring diverse spatial discretizations and numerical algorithms: the rotating‐sphere shallow water model ShallowWaters.jl, the finite‐volume ocean model Oceananigans.jl, the finite‐element ice sheet model DJUICE.jl, and the spectral atmospheric model SpeedyWeather.jl. Across these ESM components, our tools compute efficient and correct gradients. These results establish a foundation for differentiable, high‐performance and performance‐portable ESMs that can integrate neural networks for unresolved processes, trained online, enabling next‐generation hybrid physics–machine learning ESMs constrained by physical dynamics and observations.


  • Open Access

    DJ4Earth: Differentiable, and Performance-portable Earth System Modeling via Program Transformations

    by William S Moses, Gong Cheng, Valentin Churavy, Maximilian Gelbrecht, Milan Klöwer, Joseph Kump, Mathieu Morlighem, Sarah Williamson, Dhruv Apte, Paul Berg, et al.

    Differentiable Earth system models (ESMs) enable powerful applications such as sensitivity analysis, gradient-based calibration, state estimation, boundary flux inversions, uncertainty quantification, and online machine learning. Reverse-mode automatic differentiation (AD) efficiently provides gradients for such tasks, yet models have rarely included this capability because of complex, bespoke numerical algorithms. As part of the \textit{Differentiable programming in Julia for Earth system modeling (DJ4Earth)} initiative, we present improved capabilities of the AD tool Enzyme.jl and the new compiler transpilation tool Reactant.jl, augmented by sophisticated checkpointing algorithms, which, together make general-purpose AD tractable and efficient for full-fledged ESM components written in Julia. Operating at the Low-Level Virtual Machine (LLVM) intermediate representation or Multi-Level Intermediate Representation (MLIR) compiler levels, these frameworks support mutable memory, custom kernels, and compiler optimizations before and after differentiation. Julia-specific challenges related to just-in-time compilation and garbage collection are handled efficiently. Reactant further enables automatic performance portability across CPUs, GPUs, and TPUs, facilitating use of emerging AI-customized high-performance computing architectures. We demonstrate these frameworks on four Julia-based ESM components featuring diverse spatial discretizations and numerical algorithms: (i) the rotating-sphere shallow water model ShallowWaters.jl, (ii) the finite-volume ocean model Oceananigans.jl, (iii) the finite-element ice sheet model DJUICE.jl, and (iv) the spectral atmospheric model SpeedyWeather.jl. Across these ESM components, our tools compute efficient and correct gradients. These results establish a foundation for differentiable, high-performance and performance-portable ESMs that can integrate neural networks for unresolved processes, trained online, enabling next-generation hybrid physics--machine learning ESMs constrained by physical dynamics and observations.

  • Open Access

    Learning Machine Learning with Lorenz-96

    by Dhruv Balwada, Ryan Abernathey, Shantanu Acharya, Alistair Adcroft, Judith Brener, V Balaji, Mohamed Aziz Bhouri, Joan Bruna, Mitch Bushuk, Will Chapman, et al.


  • Open Access

    Comparing Two Parameterizations for the Restratification Effect of Mesoscale Eddies in an Isopycnal Ocean Model

    by Nora Loose, Gustavo M. Marques, Alistair Adcroft, Scott Bachman, Stephen M. Griffies, Ian Grooms, Robert W. Hallberg, Malte F. Jansen

    There are two distinct parameterizations for the restratification effect of mesoscale eddies: the Greatbatch and Lamb (1990, GL90, https://journals.ametsoc.org/view/journals/phoc/20/10/1520-0485_1990_020_1634_opvmom_2_0_co_2.xml?tab_body=abstract-display) parameterization, which mixes horizontal momentum in the vertical, and the Gent and McWilliams (1990, GM90, https://journals.ametsoc.org/view/journals/phoc/20/1/1520-0485_1990_020_0150_imiocm_2_0_co_2.xml) parameterization, which flattens isopycnals adiabatically. Even though these two parameterizations are effectively equivalent under the assumption of quasi‐geostrophy, GL90 has been used much less than GM90, and exclusively in z‐coordinate models. In this paper, we compare the GL90 and GM90 parameterizations in an idealized isopycnal coordinate model, both from a theoretical and practical perspective. From a theoretical perspective, GL90 is more attractive than GM90 for isopycnal coordinate models because GL90 provides an interpretation that is fully consistent with thickness‐weighted isopycnal averaging, while GM90 cannot be entirely reconciled with any fully isopycnal averaging framework. From a practical perspective, the GL90 and GM90 parameterizations lead to extremely similar energy levels, flow and vertical structure, even though their energetic pathways are very different. The striking resemblance between the GL90 and GM90 simulations persists from non‐eddying through eddy‐permitting resolution. We conclude that GL90 is a promising alternative to GM90 for isopycnal coordinate models, where it is more consistent with theory, computationally more efficient, easier to implement, and numerically more stable. Assessing the applicability of GL90 in realistic global ocean simulations with hybrid coordinate schemes should be a priority for future work.


  • Diagnosing Scale-Dependent Energy Cycles in a High-Resolution Isopycnal Ocean Model

    by Nora Loose, Scott Bachman, Ian Grooms, Malte Jansen

    Energy exchanges between large-scale ocean currents and mesoscale eddies play an important role in setting the large-scale ocean circulation but are not fully captured in models. To better understand and quantify the ocean energy cycle, we apply along-isopycnal spatial filtering to output from an isopycnal 1/32° primitive equation model with idealized Atlantic and Southern Ocean geometry and topography. We diagnose the energy cycle in two frameworks: 1) a non-thickness-weighted framework, resulting in a Lorenz-like energy cycle, and 2) a thickness-weighted framework, resulting in the Bleck energy cycle. This paper shows that framework 2 is more useful for studying energy pathways when an isopycnal average is used. Next, we investigate the Bleck cycle as a function of filter scale. Baroclinic conversion generates mesoscale eddy kinetic energy over a wide range of scales and peaks near the deformation scale at high latitudes but below the deformation scale at low latitudes. Away from topography, an inverse cascade transfers kinetic energy from the mesoscales to larger scales. The upscale energy transfer peaks near the energy-containing scale at high latitudes but below the deformation scale at low latitudes. Regions downstream of topography are characterized by a downscale kinetic energy transfer, in which mesoscale eddies are generated through barotropic instability. The scale- and flow-dependent energy pathways diagnosed in this paper provide a basis for evaluating and developing scale- and flow-aware mesoscale eddy parameterizations. Significance Statement: Blowing winds provide a major energy source for the large-scale ocean circulation. A substantial fraction of this energy is converted to smaller-scale eddies, which swirl through the ocean as sea cyclones. Ocean turbulence causes these eddies to transfer part of their energy back to the large-scale ocean currents. This ocean energy cycle is not fully simulated in numerical models, but it plays an important role in transporting heat, carbon, and nutrients throughout the world’s oceans. The purpose of this study is to quantify the ocean energy cycle by using fine-scale idealized numerical simulations of the Atlantic and Southern Oceans. Our results provide a basis for how to include unrepresented energy exchanges in coarse global climate models.


  • Open Access

    Parameterizing eddy form stress in a thickness-weighted average isopycnal ocean model

    by Nora Loose, Gustavo M Marques, Alistair Adcroft, Scott Daniel Bachman, Stephen M Griffies, Ian Grooms, Robert W. Hallberg, Malte Friedrich Jansen

    Mesoscale eddies are unresolved in most of today's global ocean models, and their effect on the large-scale ocean circulation needs to be parameterized. Greatbatch and Lamb (1990, GL90) suggested a...

  • Open Access

    Diagnosing scale-dependent energy cycles in a high-resolution isopycnal ocean model

    by Nora Loose, Scott Bachman, Ian Grooms, Malte Jansen

    Energy exchanges between large-scale ocean currents and mesoscale eddies play an important role in setting the large-scale ocean circulation but are not fully captured in models. To better understand and quantify the ocean energy cycle, we apply along-isopycnal spatial filtering to output from an isopycnal 1/32$^\circ$ primitive equation model with idealized Atlantic and Southern Ocean geometry and topography. We diagnose the energy cycle in two frameworks: (1) a non-thickness-weighted framework, resulting in a Lorenz-like energy cycle, and (2) a thickness-weighted framework, resulting in the Bleck energy cycle. This paper shows that (2) is the more useful framework for studying energy pathways when an isopycnal average is used. Next, we investigate the Bleck cycle as a function of filter scale. Baroclinic conversion generates mesoscale eddy kinetic energy over a wide range of scales, and peaks near the deformation scale at high latitudes, but below the deformation scale at low latitudes. Away from topography, an inverse cascade transfers kinetic energy from the mesoscales to larger scales. The upscale energy transfer peaks near the energy-containing scale at high latitudes, but below the deformation scale at low latitudes. Regions downstream of topography are characterized by a downscale kinetic energy transfer, in which mesoscale eddies are generated through barotropic instability. The scale- and flow-dependent energy pathways diagnosed in this paper provide a basis for evaluating and developing scale- and flow-aware mesoscale eddy parameterizations.