Search NASA⌕ Search

SEARCH · Search NASA

Results for “Model transferability”

Search indexed NASA NTRS and DOE OSTI research on propulsion, heat transfer, battery materials and energy systems. Follow report and document links to the original sources.

Quote a phrase for an exact phrase match. Source license links do not imply unrestricted reuse.

At least 289 records · Page 16
All sourcesNASA NTRSDOE OSTI
DOE OSTI2026

Foundation models for atomistic simulation of chemistry and materials

Conventional computational methods for modeling chemical and materials systems are limited by system size and timescale, forcing a trade-off between quantum-mechanical accuracy and the sampling needed for realistic observables. Large language and vision foundation models — pre-trained on massive datasets using transformer architectures — have revolutionized many fields. It is thus interesting to ask whether a foundation model — subject to suitable data, parameter scaling and training — could enable learned simulations of chemistry and materials. Here, in this study, we review the field of machine-learned interatomic potentials (MLIPs) and posit that scaling up large and diverse chemical and materials datasets and highly expressive architectures using advanced training strategies should result in models that are: more efficient, transferable, robust to out-of-distribution scenarios, and easier to fine-tune to a variety of downstream physical observables than models trained from scratch on small datasets corresponding to specific, targeted atomistic simulation tasks. We provide specific criteria for creating such large-scale MLIP foundation models, coordinated strategies for their development, evaluation and deployment, and highlight potential emergent capabilities that could transform predictive simulations in chemistry and materials science and accelerate discovery across multiple technological domains.

Yuan, Eric C.-Y. [University of California, Berkel↗
DOE OSTI2026

Hierarchical transfer learning: an agile and equitable strategy for machine-learning interatomic models

Machine-learned interatomic models are growing in popularity due to their ability to afford near quantum-accurate predictions for complex phenomena with orders-of-magnitude greater computational efficiency. However, these models struggle when applied to systems of many element types due to the approximately exponential increase in number of parameters that must be determined. To mitigate this challenge, we present a new hierarchical transfer learning approach that allows the fitting problem to be decomposed into smaller independent and reusable parameter blocks that enable development of explicitly chemically extensible ML-IAM. Application of this strategy is demonstrated for C and N mixtures under conditions ranging from nominally ambient to ~10,000 K and 200 GPa for compositions from 0 to 100% N. Ultimately, this strategy makes model generation for chemically complex systems more tractable and efficient, facilitates comprehensive model validation, and makes ML-IAM development for problems of this nature more accessible to users with limited access to extreme computing infrastructure.

Lindsey, Rebecca K. [Univ. of Michigan, Ann Arbor,↗
DOE OSTI2024

Development & Experimental Validation of a Generalized Resistance-Capacitance Model for Numerical Simulation of Phase-Change Material Embedded Heat Exchangers

Latent heat thermal energy storage (LHTES) using phase change material (PCM) has attracted increased attention as a viable solution for overcoming the mismatch between energy supply and demand for renewable energy-based systems. PCM-embedded heat exchangers (PCM-HX) have the potential to significantly improve thermal performance due to high storage capacity and low temperature variation during the phase change process. Most models for simulating LHTES heat transfer use Computational Fluid Dynamics (CFD) simulations, which have high computational costs resulting from considering the complex and time-dependent physics relevant to PCM-HXs. In this paper, a Generalized Resistance Capacitance-based Model (GRCM) was developed to predict the thermal performance of arbitrary PCM-HXs in a computationally efficient manner without compromising modeling accuracy. The GRCM is exercised for three case studies: (i) verification for a single-slabbed finned PCM-HX, (ii) verification and validation for a copper foam/paraffin composite PCM-HX, and (iii) validation for a straight tube annular finned PCM-HX. The copper foam PCM-HX uses an electric heater at the top of HX, while the other two configurations utilize water as heat transfer fluid. For the single-slabbed finned PCM-HX melting case, the mean deviation in average PCM temperature predicted by the GRCM compared to the CFD model was between 0.56 – 0.73 K, with maximum temperature deviation of 2.68 K. For the HTF outlet temperature, the validation results showed that GRCM prediction matches very well with experimental data, with mean temperature deviation of 0.24 K during melting case, while for solidification case was 0.34 K. These results showcase the GRCM’s capability for accurately reproducing the thermal characteristics of PCM-HXs with considerably lower computational effort.

42 ENGINEERING↗
DOE OSTI2024

Intermolecular Proton Transfer Enabled Reactive CO 2 Capture by the Malononitrile Anion

Task-specific ionic liquids (ILs) employing carbanions represent a new class of ILs for carbon capture. The deprotonated malononitrile carbanion, [CH(CN) 2 ] - , has shown close to equimolar capacity for reactive CO 2 capture. Although the formation of the [C(CN) 2 COOH] - carboxylic acid was found to be the final product, how the hydrogen atom on the [CH(CN) 2 ] - carbanion transfers to the carboxylate group as a proton has not been fully understood. In this work, we employ density functional theory calculations with an implicit solvation model to investigate the proton transfer mechanisms in forming carboxylic acid from the reaction of the [CH(CN) 2 ] - carbanion with CO 2 . We find that the intramolecular proton-transfer pathway in [CH(CN) 2 COO] - to form [C(CN) 2 COOH] - is unlikely due to the high energy barrier of 152 kJ/mol. Instead, the intermolecular proton transfer pathway between two [CH(CN) 2 COO] - anions is more feasible to form two molecules of [C(CN) 2 COOH] - , with a significantly lower activation energy of 50 kJ/mol. Moreover, the [C(CN) 2 COOH] - dimer is further stabilized by the intermolecular hydrogen bonds of the two –COOH groups in the Z-configuration of the π-conjugated planar geometry. This insight of reactive CO 2 capture enabled by intermolecular proton transfer will be useful in designing novel carbanions and ILs for carbon capture and conversion.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗
DOE OSTI2026

JANUS: Resilient and Adaptive Data Transmission for Enabling Timely and Efficient Cross-Facility Scientific Workflows

In modern science, the growing complexity of large-scale scientific projects has led to an increasing reliance on cross-facility scientific workflows, where resources and expertise from multiple institutions and geographic locations are leveraged to accelerate scientific discovery. These workflows often require transmitting huge amounts of scientific data through wide-area networks. Although high-speed networks like ESnet and transfer services such as Globus have improved data mobility, several challenges remain. The sheer volume of data can overwhelm network bandwidth, widely used transport protocols such as TCP suffer from inefficiencies due to retransmissions triggered by packet loss, and existing fault-tolerance mechanisms like erasure coding introduce substantial overhead. In this paper, we propose Janus, a resilient and adaptable data transmission approach designed for cross-facility scientific workflows. Unlike traditional TCP-based methods, Janus leverages UDP, integrates erasure coding for fault tolerance, and combines it with error-bounded lossy compression to reduce overhead. This novel design allows users to balance data transmission time and accuracy, optimizing transfer performance based on specific scientific requirements. Additionally, Janus dynamically adjusts erasure coding parameters in response to real-time network conditions, ensuring efficient data transfers even in fluctuating environments. We develop optimization models for determining ideal configurations and implement adaptive data transfer protocols to enhance reliability. Through extensive simulations and real-network experiments, we demonstrate that Janus significantly improves transfer efficiency while maintaining data fidelity.

Esaulov, Vladislav [Georgia State University, Atla↗
DOE OSTI2024

Analysis of heat and mass transfer potential of a dew-point cooling tower in different climatic conditions

In this study, the performance of the Dew-Point Cooling Tower (DPCT) was analyzed for different factors in a variety of climate conditions. For this purpose, a dedicated numerical model describing heat and mass transfer processes was developed and validated. The results of the numerical simulations allowed to analyze the potential of utilizing the heat and mass transfer process with the dew-point phenomenon for water cooling. It was established that the operational parameters that have a high impact on the performance of the DPCT are: inlet water temperature and inlet air humidity ratio. It was also established that DPCT achieves the highest COP and Specific Cooling Capacity for cold subtropical highland climates and that it achieves highest Wet-bulb Effectiveness for monsoon-influenced humid subtropical climates. In conclusion, the regions where all the efficiency factors achieved above-average values included warm, arid, and desert climates.

42 ENGINEERING↗
DOE OSTI2025

High-Speed and High-Quality Field Welding Repair Based on Advanced Non-Destructive Evaluation and Numerical Modeling

Creep strength-enhanced ferritic (CSEF) steels such as Grade 91 (9Cr-1Mo-V) and Grade 92 (Fe-9Cr-2W-0.5Mo) steels are widely used in the fossil-fuel-fired and nuclear power plants. The weld integrity of these steels is crucial for power plants' safe and reliable operations. Due to harsh service conditions, the steel weld can become susceptible to environmental degradation. Field welding repair is used to restore the degraded weld’s performance where a controlled temper-bead welding technique is commonly used to temper the freshly formed martensite during welding. However, knowledge of weld repairability is limited and experimental trial and error optimization to achieve desired microstructure and joint properties is expensive and time-consuming. Many existing computational models, e.g., finite element models, are limited to solving heat conduction equation and ignoring convective heat transfer due to molten metal flow. These models can result in over-prediction of peak temperatures of weld pool and heat-affected zone (HAZ), which in turn can affect the accuracy of tempering prediction. Moreover, these finite element models require an input of the deposit profiles in advance and thus limits the usability of these models. Here, a molten pool-based, multi-pass multi-layer model has been developed based on computational fluid dynamics (CFD) approach with the Volume of Fluid (VOF) method. The model calculates the bead formation, thereby eliminating the need for pre-determined bead profiles required by finite element models. For computational efficiency, a coordinate system attached to the moving heat source is utilized. A subroutine is developed to convert the temperature profiles in the reference frame stationary to the heat source to that stationary to the workpiece. The converted thermal cycles are then imported into a microstructure model to compute the tempering kinetics and resultant hardness using a Johnson-Mehl-Avrami-Kolmogorov (JMAK), and modified Grange-Baughman parameter. The modeling approach is first developed and validated on single- and multi-pass deposition of stainless steel filler metal onto a SA-533 high strength steel substrate. The models are then applied to a multi-pass V-groove repair weld of Grade 91 steel plate as well as directed energy deposition of Grade 92 steel. Non-destructive characterization of microstructures was performed on Grade 91 and 92 steel welds. Two welding processes, cold metal transfer (CMT) and flux-cored arc welding (FCAW), were investigated for the Grade 91 steel weld samples. For the Grade 92 weld samples, three different heat inputs (low, medium, and high) of gas tungsten arc welding (GTAW) were utilized to replicate traditional field welding processes. The non-destructive evaluation (NDE) method used for this research was immersion ultrasonic testing (UT) using a micro-resolution ultrasonic imaging methodology specifically designed to operate in the through-transmission configuration operating at 20 MHz of frequency. The system used a focused ultrasonic beam spot size diameter between 250-300 μm, and a 6 μm laser vibrometer spot size for detection, to produce highly defined images with longitudinal and mode-converted shear waves. From the micro-resolution ultrasonic C-scan images, three microstructural regions, i.e., weld metal (WM), HAZ, and base metal (BM), were clearly identifiable. Various levels of ultrasonic amplitudes distributed over the three regions were correlated with electron beam backscattered diffraction (EBSD) images using grain size, grain boundaries, and dislocation densities. The results showed that areas with relatively higher ultrasonic amplitude levels were associated with smaller grains and higher dislocation densities, while areas with lower amplitude levels were associated with larger grains and lower dislocation densities. In addition, ultrasonic velocity data obtained across the three different weld microstructural regions of Grade 91 test samples were correlated with optical metallographic images and hardness measurements. The results showed distinctive decreases in ultrasonic velocity and hardness over the HAZ region, where weld failures often occur during service.

36 MATERIALS SCIENCE↗
DOE OSTI2025

Photoelectrochemical Hydride Generation with Oxide-Coated Silicon

Photoelectrochemical generation of a potent organic hydride donor at silicon is demonstrated. Two different oxide-coated p-type silicon photoelectrodes reduced 1,2,3,5,6-pentamethyl-1H-benzo[d]imidazol-3-ium hexafluorophosphate, [PMBI][PF 6 ], to its corresponding imidazole hydride, PMBIH, in the presence of a proton source. Under 1 sun illumination, in acetonitrile with 1,5-diazabicyclo[4.3.0]non-5-ene (DBN) buffer, the p-Si photoelectrodes convert PMBI + to PMBIH with good Faradaic efficiencies (FEs): 78% FE at −2.3 V vs Fc +/0 for Si|TiO 2 and 83% FE at −2.6 V vs Fc +/0 for Si|SiO 2 (where Si|SiO 2 represents silicon coated with an oxide layer). Generally, the Si|TiO 2 catalyzed the reaction at milder potentials than Si|SiO 2 , but the Si|SiO 2 had better selectivity for PMBIH generation over H 2 evolution than Si|TiO 2 . In light of prior studies of these photoelectrodes, the differences in selectivity and onset potential suggest a difference in mechanism, likely an outer-sphere electron transfer (ET) mechanism at the SiO 2 surface and potentially a proton-coupled ET process at the TiO 2 surface. To help understand reaction efficiency and identify areas of improvement, a thermochemical model for understanding net hydride transfer from the semiconductor to an acceptor in solution was developed. We find that the reactions in the present system are quite downhill. The high overpotentials (even while maintaining selectivity over H 2 evolution) emphasize the need for improved catalysts. Furthermore, this approach to evaluate the thermodynamics of net hydride transfer should be broadly valuable for electrochemical and photoelectrochemical processes.

14 SOLAR ENERGY↗
DOE OSTI2024

Mechanistic definition and prediction of the mass exchange coefficient between rivers and hyporheic zones: The $α$ of two $Ω$s

Solute transport in interconnected rivers and hyporheic zones is typically modeled through dual-domain models where first-order solute mass transfer between the two domains, Ω R and Ω HZ , is represented by a coefficient α. The transient storage model (TSM) is an example of such an approach. In practice, α is determined by fitting the tails of solute tracer breakthrough curves using a TSM. This approach has led to ambiguity regarding α’s physical meaning and transferability. Here, in this work, we investigated the physical basis for α and tested it with virtual experiments through the fully coupled multiphysics model hyporheicFoam for the Ω R – Ω HZ system. hyporheicFoam explicitly simulated coupled flow and solute transport over a kilometer with centimeter-scale resolution. Model results were analyzed to calculate α following its theoretical definition directly. Using the determined α within a TSM enables accurate reproduction of solute transport, underscoring α’s physical relevance and precision.

54 ENVIRONMENTAL SCIENCES↗
DOE OSTI2025

Transfer learning nonlinear plasma dynamic transitions in low dimensional embeddings via deep neural networks

Deep learning algorithms provide a new paradigm to study high-dimensional dynamical behaviors, such as those in fusion plasma systems. Development of novel, data-driven model reduction methods, coupled with detection of abnormal modes with plasma physics, opens a unique opportunity to identify plasma instabilities through automated construction of parsimonious models that can be tuned to balance accuracy and cost. Our fusion transfer learning (FTL) model demonstrates success in rapidly reconstructing nonlinear kink mode structures by learning from a limited amount of nonlinear simulation data. The knowledge transfer process leverages a pre-trained neural encoder–decoder network, initially trained on linear simulations, to effectively capture nonlinear dynamics. The low-dimensional embeddings extract the coherent structures of interest, while preserving the inherent dynamics of the complex system. Experimental results highlight FTL’s capacity to capture transitional behaviors and dynamical features in plasma dynamics—a task often challenging for conventional methods. The model developed in this study is generalizable and can be extended broadly through transfer learning to address various magnetohydrodynamics modes.

70 PLASMA PHYSICS AND FUSION TECHNOLOGY↗
DOE OSTI2024

Hiding-in-Plain-Sight (HiPS) Attack on CLIP for Targetted Object Removal from Images

Machine learning models are known to be vulnerable to adversarial attacks, but prior works have mostly focused on single-modalities. With the rise of large multi-modal models (LMMs) like CLIP, which combine vision and language capabilities, new vulnerabilities have emerged. However, these multimodal targeted attacks aim to completely change the model's output to what the adversary wants. In many realistic scenarios, an adversary might seek to make only subtle modifications to the output, so that the changes go unnoticed by downstream models or even by humans. We introduce Hiding-in-Plain-Sight (HiPS) attacks, a novel class of adversarial attacks that subtly modifies model predictions by selectively concealing target object(s), as if the target object was absent from the scene. We propose two HiPS attack variants, HiPS-cls and HiPS-cap, and demonstrate their effectiveness in transferring to downstream image captioning models, such as CLIP-Cap, for targeted object removal from image captions.

Daw, Arka [ORNL] (ORCID:0009000633191271)↗
DOE OSTI2025

Optimal Design and Operation of Intensified Absorbers with 3D-Printed Packing for Solvent-Based CO 2 Capture

Many potential solvent-based carbon capture processes suffer from a high heat of absorption of CO 2 that adversely affects the thermodynamic driving force. While interstage coolers are often used for removing a portion of the generated heat by removing the solvent or a portion of the solvent from a stage and cooling and returning it back to the absorber, they can be placed only at discrete locations in the tower. This work investigates intensified absorbers with 3D-printed packing that includes an internal cooler and therefore can be potentially used for maximizing the operational efficiency of the absorbers for CO 2 capture. The intensified absorber is modeled by using a generic, first-principles, equation-oriented absorber column model. Since the placement of these intensified packings would cause a loss of area/volume used for mass transfer, optimization of the proposed intensified absorber is performed by optimally selecting the locations at which to place these devices and designing them such that the trade-off due to the addition of the heat removal area and the resulting loss in the mass transfer area is accounted for. Results show that optimally placed and designed intensified packings can lead to a significant increase in the capture efficiency of the process in comparison to a similar column with no internal cooling. It is also observed that by optimally placing and designing intensified packings, the lean solvent flow rate to the absorber can be decreased, and the CO 2 lean loading can be increased while still maintaining the same capture efficiency. These process changes can lead to a substantial reduction in the cost of capture.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗
DOE OSTI2025

Machine learning for reactor power monitoring with limited labeled data

Real-time reactor power monitoring is critical for a variety of nuclear applications, spanning safety, security, operations, and maintenance. While machine learning methods have shown promise in monitoring reactor power levels, there is limited research on their efficacy in label-starved environments. The goal of this work is to assess the feasibility of classifying nuclear reactor power level using multisource data in scenarios with limited labels. Data were collected using low-resolution multisensors at four nuclear reactor facilities: two large research reactors and two TRIGA reactors. Within each pair, one reactor dataset served as the source and the other as the target in a transfer learning paradigm. Twenty-three supervised models were trained on labeled sequences of magnetic field and acceleration data from each of the target sites. Self-learning and transfer learning methods were applied to the top performing models to assess their classification performance with increasing amounts of labeled data. While reactor power level classification was achieved with a Matthews Correlation Coefficient of up to 0.739 ± 0.003 and 0.622 ± 0.009 with only 400 sequences per power state for the large research reactor and TRIGA target sites, respectively, self-learning and transfer learning leveraging source site data did not improve target classification performance. These findings suggest that alternative methods, such as higher sensitivity sensors, digital twins, or the use of physics-informed models, are required to enable high-performance classification in machine learning approaches to reactor monitoring with a dearth of target ground truth.

46 INSTRUMENTATION RELATED TO NUCLEAR SCIENCE AND ↗
DOE OSTI2026

Thermal Performance of Spandrel Assemblies in Glazed Wall Systems: Laboratory Test Design – Challenges and Test Results

Accurate thermal performance calculation procedures for opaque spandrel areas in curtain wall and window wall systems are essential for rating systems when comparing spandrel systems. However, there is a lack of consensus in thermal modeling needed for accurately characterizing heat transfer through spandrel assemblies due to the complex arrangement of materials and structural components. Several studies indicate that conventional 2D thermal simulations may overestimate R-values by 30% compared to physical testing and 3D simulations. Detailed simulations and well-curated laboratory test data are necessary to build confidence in simulation models, which will later be used to develop correlations to improve widely used conventional 2D thermal simulations. This study aims to experimentally test heat transfer through various spandrel assemblies to validate 3D simulation models. Also, the challenges of conducting a thorough testing design along with the solutions would be documented. The team developed a design for testing spandrel assemblies, making appropriate modifications to the existing heat, air, and moisture (HAM) chamber to accommodate the testing needs. Two moveable baffles were designed and fabricated to guide airflow direction parallel to the test article surface. The data acquisition capabilities in the chamber were upgraded to add more than two hundred sensors to the climate and indoor side of the chamber. The goal is to provide a quality dataset for validating complex 3D modeling simulations, which will be used to develop improved thermal simulation techniques that more accurately represent the thermal behavior of spandrel assemblies and their integration within the building envelope. This paper will summarize the results for the boundary conditions of the testing and the temperature variation across different locations of the spandrel assemblies.

Kunwar, Niraj [ORNL] (ORCID:0000000263457652)↗
DOE OSTI2026

Hybrid Oscillator-Qubit Quantum Processors: Instruction Set Architectures, Abstract Machine Models, and Applications

This tutorial offers a pedagogical guide to hybrid quantum processors that integrate discrete-variable (DV) qubits and continuous-variable (CV) oscillators. Aimed at computer scientists, engineers, and physicists, it provides an overview of the experimental, algorithmic, and architectural aspects of this novel and rapidly developing hardware model. Experimental realizations of this model include superconducting, trapped-ion, and neutral-atom platforms. By combining DV and CV components, hybrid oscillator-qubit processors enable a powerful new paradigm that offers complementary strengths for quantum control, error correction, computation, and simulation. Working toward the goal of a full-stack system connecting applications to CV-DV hardware, we define and formulate abstract machine models and instruction set architectures. These essential abstractions enable codesign of hardware and software, and resource estimation for exploring the potential of current and future hardware for computational and simulation tasks. Using these abstractions, we present both new and existing examples that illustrate the benefits of hybrid CV-DV processors relative to traditional DV-only hardware in computation as well as quantum simulation of physical models. Examples include algorithms for transferring states between DV and CV systems, performing the quantum Fourier transform, and simulation of lattice gauge theories. Relative to qubit-only hardware, the bosonic degrees of freedom natively available in hybrid architectures can substantially reduce the circuit complexity of simulations for physical models containing bosons. A key technique is the extension of quantum signal processing ideas to CV-DV systems. This work is intended to serve as a timely and comprehensive guide to this relatively unexplored yet promising approach to quantum computation and to provide a road map to guide future development.

73 NUCLEAR PHYSICS AND RADIATION PHYSICS↗
DOE OSTI2024

Modelling of cracking during beam oscillation laser welding of a creep-resistant nickel alloy

During high power laser welding of Inconel 740H, horizontal solidification cracks form at locations between approximately 70–80% of the weld depth. With the integration of circular beam oscillation, the laser energy density distribution and underlying thermo-mechanical conditions were altered. By coupling three-dimensional heat transfer, fluid flow, and stress modelling tools, the role that circular beam oscillation plays in the formation of the strain rates and stresses driving this cracking phenomenon was identified. While the addition of a circular oscillation pattern appeared to lower the stress levels along the solidification front, it did not eliminate the appearance of horizontal cracking. Only when the beam amplitudes reached 1.6 mm did solidification cracking disappear, but the weld depths were also significantly reduced.

Gao, Mingze↗
DOE OSTI2025

Size-Transferable Prediction of Excited State Properties for Molecular Assemblies with a Machine Learning Exciton Model

Computational modeling of the excited states of molecular aggregates faces significant computational challenges and size heterogeneity. Current machine learning (ML) models, typically trained on specific-sized aggregates, struggle with scalability. We found that the exciton model Hamiltonian of large aggregates can be decomposed into dimer pairs, allowing an ML model trained on dimers to reconstruct Hamiltonians for aggregates of any size. We also proposed a new method to address the phase-correction problem by introducing coupling terms’ approximations. Our model accurately predicted the excitation energies of the trimer and tetramer of perylene and tetracene and estimated S1 oscillator strengths of perylene aggregates. Leveraging our ML model, the optical gaps of nanosized perylene aggregates with up to 50 monomers are analyzed, qualitatively revealing the role of different couplings on their size dependency. Future work will explore transferability across different monomers to predict optical properties in heterogeneous assemblies.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗
DOE OSTI2026

NLML: A Deep Neural Network Emulator for the Exact Nonlinear Interactions in a Wind Wave Model

Nonlinear wave interactions describe the resonant energy transfer between wave components, playing a fundamental role in the evolution of ocean wave spectra. Nonlinear wave interactions significantly influence wave growth and development, making them essential for accurate wave modeling. However, resolving the full six-dimensional Boltzmann integral of the exact nonlinear wave interactions (Webb-Resio-Tracy method, WRT) is computationally expensive, limiting its application in real-time operational wave forecasting and for research purposes. Current approximations, such as the Discrete Interaction Approximation (DIA), prioritize computational speed over accuracy, resulting in significant errors in wave mean parameters. Here, we introduce NLML, a machine learning (ML) emulator designed to approximate the exact nonlinear wave interactions within WAVEWATCH III (WW3), with the goal of achieving the accuracy of WRT while maintaining the stability and computational speed of DIA. By leveraging GPU capabilities such as half precision inference, we achieved substantial speedups, up to 136x mathematical equation faster than the WRT and only a modest 1.04x mathematical equation slowdown relative to DIA, while achieving 2x mathematical equation the accuracy of DIA in global wave spectral energy and mean wave parameters, with up to 7x mathematical equation higher accuracy in some regions. Unlike previous ML approaches, NLML maintained inherent stability throughout model integration in a standalone, year-long WW3 simulation, without requiring additional constraints. Our new ML parameterization bridges the gap between accuracy and efficiency, offering a promising alternative for improving wave modeling in operational settings and research purposes.

16 TIDAL AND WAVE POWER↗