Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “language models”

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 667 records · Page 37
All sourcesNASA NTRSDOE OSTI
DOE OSTI2025

Autonomous Operations for Advanced Reactors Utilizing Supervisory Control

Automation is a critical tenet of reactor plant operations as reliance on nuclear energy increases. Nuclear power plants require a large workforce which does not scale with output; that is, the cost per megawatt increases as reactor output becomes smaller. The economic viability of advanced reactors, particularly small modular reactors (SMRs) and microreactors, requires a significantly reduced onsite workforce. The logical solution is establishing a systematic process of elimination of reliance on human operators, and to the extent possible, replacing these actions with automated functions. In this paper, we propose a method for such transformation to establish a robust technical basis to enable transition to autonomy. Our method is based on finite state automata (FSA)—also known as finite state machines (FSMs). Relying on this method allows us to exploit the rich set of mathematical proofs available in the field of regular languages. FSA are one of the mathematical tools to model discrete event systems (DES). These properties are applied to produce an automated startup controller for the Massachusetts Institute of Technology Research Reactor (MITR). The startup procedure is captured in terms of discrete changes from one state to another while an independent supervisory control system directs the sequence of states and alerts a human in the event of an abnormal operation. First, the design and behavior of the MITR rod control system were modeled in Simulink. Then, the startup procedure was applied to the rod control system and the DES performed a startup by procedurally withdrawing rods to the subcritical position. The simulation also stops rod motion in response to an uncontrollable event and restarts rod motion once the event has been cleared.

46 - INSTRUMENTATION RELATED TO NUCLEAR SCIENCE AN↗
DOE OSTI2025

JACC.shared: Leveraging HPC Metaprogramming and Performance Portability for Computations That Use Shared Memory GPUs

In this work, we present JACC.shared, a new feature of Julia for ACCelerators (JACC), which is the performanceportable and metaprogramming model of the just-in-time and LLVM-based Julia language. This new feature allows JACC applications to leverage the high-performance computing (HPC) capabilities of high-bandwidth, on-chip GPU memory. Historically, exploiting high-bandwidth, shared-memory GPUs has not been a priority for high-level programming solutions. JACC.shared covers that gap for the first time, thereby providing a highlevel, portable, and easy-to-use solution for programmers to exploit this memory and supporting all current major accelerator architectures. Well-known HPC and AI workloads, such as multi/hyperspectral imaging and AI convolutions, have been used to evaluate JACC.shared on two exascale GPU architectures hosted by some of the most powerful US Department of Energy supercomputers: Perlmutter (NVIDIA A100) and Frontier (AMD MI250X). The performance evaluation reports speedup of up to 3.5× by adding only one line of code to the base codes, thus providing important accelerators in a simple, portable, and transparent way and elevating the programming productivity and performance-portability capabilities for Julia/JACC HPC, AI, and scientific applications.

Valero Lara, Pedro [ORNL] (ORCID:0000000214794310)↗
DOE OSTI2024

Innovating the next generation of commercial smart building software

Nearly 30% of commercial building energy use is wasted due to equipment faults and HVAC controls problems. The result is increased emissions, compromised comfort and productivity, and less reliable coordination of building power needs with a clean grid. The energy impact alone represents $17 billion in potential savings. Today’s smart building software provides a robust solution to address these operational deficiencies. Energy management and information systems (EMIS) are saving up to 9% on average, with two-year paybacks. They are being incorporated into energy management processes, commissioning services, and utility programs. As effective as they are, two barriers prevent even deeper benefits; limited personnel to fix problems once they are identified, and the expense and time to manually implement changes in control systems. In partnership with the research community, the EMIS industry is developing new capabilities to overcome these barriers. Moving beyond siloed products for either fault detection and diagnostics, or optimal control, these new capabilities empower users to not only automatically identify faults, but also to push corrective action, and control improvements to their buildings. In this paper, several areas for enhancements are documented: ‘one-time’ correction of faults such as setpoints, schedules, and economizer lockouts; short-term active testing for automated proportional integral derivative (PID) loop tuning and functional testing; and continuous supervisory control for demand flexibility and year-round efficiency. Results are presented from a pair of partner implementations out of a dozen providers integrating these enhancements into their products, including field tests from across the country, and insights into operator acceptance and integration into operations and maintenance practices.

Casillas, Armando↗
DOE OSTI2021

Open Building Control

Best practice control sequences are often not implemented correctly, or are not implemented at all, in large commercial buildings. This typically leads to 10-30 percent energy waste, along with reduced occupant productivity and unnecessary equipment wear. The current process of designing and implementing such control sequences is a manual process that starts with designers who often don’t have adequate training, then requires controls programmers to interpret and program a verbose written sequence. This process has been shown to fail to deliver high performance control sequences at scale. The Open Building Control project digitizes the current control delivery process. The project is developing tools for system designers to select control sequences, assess their energy performance and load flexibility potential using whole building simulation, specify the sequence for implementation using machine-to-machine translation by a control provider and formally testing the as-installed sequences by a commissioning agent. The project developed tools for each stage of this delivery process. The key innovation of the project is the development of the Control Description Language, a language that allows such a digitized control delivery process with end-to-end verification. Libraries of control sequences have been implemented using the Control Description Language, and their performance has been demonstrated using whole building energy simulation. An automated translation of such sequences to a commercial control product line has been conducted using a prototype translator. Tools for formal verification of as-installed control sequences relative to their specification have been developed and demonstrated. The American Society of Heating, Refrigerating and Air-Conditioning Engineers (ASHRAE) started the process of forming a committee to make this language an ASHRAE/ANSI Standard. This new standard will complement existing and emerging ASHRAE standards for building communication and semantic modeling by providing a standard for expressing the control logic - the actual brain of the building. We expect this language and the process it enables to be an important contribution to the deployment of high performance building control sequences at scale because it allows taming the complexity of the control delivery process, which is continually increasing due to the need for higher performance and increased load flexibility to meet goals for net zero energy and increased renewable integration.

32 ENERGY CONSERVATION, CONSUMPTION, AND UTILIZATI↗
DOE OSTI2025

Comprehensive Database of Environmental Mitigations Extracted from FERC-Licensed Hydropower Projects Using Artificial Intelligence Techniques, 1998-2023

This dataset provides a comprehensive inventory of environmental mitigation measures required by Federal Energy Regulatory Commission (FERC) licensed hydropower facilities from 461 licenses that were issued from 1998 to 2023. These licenses constitute 446 of the 1015 FERC projects that were active at the end of 2023. 17,612 mentions of environmental mitigations were identified and categorized in 128 unique categories. Mitigations were identified using a Natural Language Processing (NLP) approach, specifically with a Bidirectional Encoder Representations from Transformer (BERT) model. Model-derived results were then reviewed and updated by a subject matter expert as needed. This dataset introduces important enhancements to previous efforts to inventory environmental mitigations, such as including associated license text for each mitigation, tracking the number of instances a mitigation was identified within a license, and providing improved location information. These enhancements significantly expand the dataset's utility, offering greater analytical capabilities and ensuring reproducibility. The dataset is downloadable as a zip file containing the metadata and dataset files.

Ruggles, Thomas [Oak Ridge National Laboratory (OR↗
DOE OSTI2026

Dominant Controls on Preferential Flow and Their Implications for Future Soil Water Fluxes

Abstract Soil water flow, particularly preferential flow (PF), is a critical control on hydrological and biogeochemical processes, including groundwater recharge, contaminant transport, and carbon cycling. However, it remains challenging to predict PF occurrence across large environmental gradients. Here, we developed a deep learning (DL) model to estimate event‐scale soil water flow velocity and the probability of PF occurrence using high‐frequency soil moisture and precipitation data from 33 sites across the National Ecological Observatory Network. The model demonstrated high skill in predicting the binary occurrence of PF (91% F1‐score; 85% accuracy) but the performance was limited in predicting soil water velocity ( R 2 = 0.31). We found that precipitation characteristics (duration, volume, and intensity) were the most important predictors for soil water velocity. Among the non‐precipitation event variables, sand content showed relatively high predictive skill, though differences among non‐event climate variables were generally modest. Lower sand content was associated with increased predicted soil water velocity, a finding that highlights the role of soil structure in producing more non‐uniform flow, which contrasts with traditional uniform flow models. Projecting a reduced DL model under both moderate and high‐emissions future climate scenarios (2060–2099 Representative Concentration Pathways 4.5 and 8.5), we found ∼7.3% increase under RCP4.5 and ∼15% under RCP8.5 of soil water velocities compared to the historical simulation, while modeled likelihood of PF changed little. These findings suggest climate change is not making PF more frequent, but it is making existing PF pathways more efficient with important consequences for associated nutrient and contaminant transport under climate change. Plain Language Summary Water movement in soil is critical for water quality. While often modeled as a uniform flow process, in reality water moves rapidly through cracks and burrows in what is called “preferential flow” (PF), which limits natural filtration and can transport pollutants. We developed a deep learning model, trained on data from 33 U.S. sites, to predict when and how fast this PF occurs based on precipitation, soil, and climate data. The model showed that precipitation characteristics (duration, intensity, volume) were the most important predictors of PF. Lower soil sand content/higher clay content was associated with faster water flow, likely due to clay soils forming aggregates and cracks that water moves through rather than infiltrating uniformly. Further analyses based on climate projections suggest that the speed at which PF occurs will become more rapid under future climate scenarios compared to historical simulation. This highlights the need to represent PF in soil water models when assessing future water quality. Key Points The effect of precipitation peak intensity on soil water velocities declined with increasing precipitation intensity Antecedent soil moisture failed to predict preferential flow (PF), contrasting the high predictive power of sand content Climate predictions suggest that soil water velocities through PF paths will increase ∼15% by 2099

Li, Bonan↗
DOE OSTI2021

Modeling of particle transport, neutrals and radiation in magnetically-confined plasmas with Aurora

In this work, we present Aurora, an open-source package for particle transport, neutrals and radiation modeling in magnetic confinement fusion plasmas. Aurora's modern multi-language interface enables simulations of 1.5D impurity transport within high-performance computing frameworks, particularly for the inference of particle transport coefficients. A user-friendly Python library allows simple interaction with atomic rates from the Atomic Data and Atomic Structure database as well as other sources. This enables a range of radiation predictions, both for power balance and spectroscopic analysis. We discuss here the superstaging approximation for complex ions, as a way to group charge states and reduce computational cost, demonstrating its wide applicability within the Aurora forward model and beyond. Aurora also facilitates neutral particle analysis, both from experimental spectroscopic data and other simulation codes. Leveraging Aurora's capabilities to interface SOLPS-ITER results, we demonstrate that charge exchange is unlikely to affect the total radiated power from the ITER core during high performance operation. Finally, we describe the ImpRad module in the one modeling framework for integrated task framework, developed to enable experimental analysis and transport inferences on multiple devices using Aurora.

70 PLASMA PHYSICS AND FUSION TECHNOLOGY↗
DOE OSTI2022

Towards Digital and Performance-Based Supervisory HVAC Control Delivery

Upgrading supervisory HVAC control in commercial buildings is one of the most attractive decarbonization tools at our disposal. Modern controls are software programs and can in theory be deployed at scale and with a low up-front carbon "pulse". In practice, however, control delivery is a disjointed and inefficient process, dominated by manual handoffs of imprecise English language documents. A particularly high barrier exists between control implementation and building energy modeling (BEM) which results in control sequences typically not being tested for correctness or performance before implementation. Together with industry partners, DOE and the national labs are developing an ecosystem of tools and standards that can support fully digital performance-based control delivery workflows. This paper describes this ecosystem, which consists of three mutually supportive efforts. Semantic models of buildings and their systems enable automatic configuration and installation of control software. Platform-neutral control descriptions separate control algorithms from control platforms and enable the creation of libraries of reference control implementations. Dynamic whole-building energy-control simulation that can execute physically realistic control sequences makes it possible to test and evaluate the performance of control sequences and then directly compile them for installation and execution in control systems. In addition to digitizing and streamlining project-level control delivery, these standards and related software support benchmarking of control algorithms, both rule-based and optimization-based, and help to both advance the state of the art and to implement ratings and programs that encourage the adoption of high-performance control.

building controls↗
DOE OSTI2024

Equivariant, safe and sensitive — graph networks for new physics

This study introduces a novel Graph Neural Network (GNN) architecture that leverages infrared and collinear (IRC) safety and equivariance to enhance the analysis of collider data for Beyond the Standard Model (BSM) discoveries. By integrating equivariance in the rapidity-azimuth plane with IRC-safe principles, our model significantly reduces computational overhead while ensuring theoretical consistency in identifying BSM scenarios amidst Quantum Chromodynamics backgrounds. The proposed GNN architecture demonstrates superior performance in tagging semi-visible jets, highlighting its potential as a robust tool for advancing BSM search strategies at high-energy colliders.

72 PHYSICS OF ELEMENTARY PARTICLES AND FIELDS↗
DOE OSTI2020

A large-scale, agent-based simulation of metropolitan freight movements with passenger and freight market interactions

This study summarizes the first stage in the implementation of an agent-based freight modeling system that has a global representation of agents and detailed modeling of a large-scale transportation network. The model is used to evaluate the transportation and energy impacts of goods movement across urban and national scales. The framework is implemented within POLARIS, a C++-based Planning and Operations Language for Agent-based Regional Integrated Simulation, which consists of an activity-based modeling (ABM) and dynamic traffic assignment (DTA) system that has robust features for passenger travel. This platform provides a tool to model interactions among consumers, producers, and the transportation system. The main objective of this initial implementation is to implement a freight model within POLARIS following an agent-based paradigm with behavioral and simulation methods. This paper presents the initial framework and illustrates the application of the model. Building upon earlier works, a parcel location assignment algorithm for business establishments in the population is documented, along with a method for estimating establishment production and consumption volumes. In addition to population generation, other features of the model include push-pull supply chains, multimodal path choice, choice of transportation logistics node, and dynamic traffic assignment. A module with e-commerce supply and demand was also developed to analyze the effects of e-commerce delivery on last-mile energy use and congestion.

agent-based freight↗
DOE OSTI2020

Low predictability of energy balance traits and leaf temperature metrics in desert, montane and alpine plant communities

Abstract Leaf energy balance may influence plant performance and community composition. While biophysical theory can link leaf energy balance to many traits and environment variables, predicting leaf temperature and key driver traits with incomplete parameterizations remains challenging. Predicting thermal offsets ( δ , T leaf − T air difference) or thermal coupling strengths ( β , T leaf vs. T air slope) is challenging. We ask: (a) whether environmental gradients predict variation in energy balance traits (absorptance, leaf angle, stomatal distribution, maximum stomatal conductance, leaf area, leaf height); (b) whether commonly measured leaf functional traits (dry matter content, mass per area, nitrogen fraction, δ 13 C, height above ground) predict energy balance traits; and (c) how traits and environmental variables predict δ and β among species. We address these questions with diurnal measurements of 41 species co‐occurring along a 1,100 m elevation gradient spanning desert to alpine biomes. We show that (a) energy balance traits are only weakly associated with environmental gradients and (b) are not well predicted by common functional traits. We also show that (c) δ and β can be partially approximated using interactions among site environment and traits, with a much larger role for environment than traits. The heterogeneity in leaf temperature metrics and energy balance traits challenges larger‐scale predictive models of plant performance under environmental change. A free Plain Language Summary can be found within the Supporting Information of this article.

Blonder, Benjamin↗
DOE OSTI2023

Practical Guide to Chemometric Analysis of Optical Spectroscopic Data

The methodology and mathematical treatment of several classic multivariate methods for the analysis of spectroscopic data is demonstrated in a straightforward way that can be used as a basis for teaching an undergraduate introductory course on chemometric analysis. The multivariate techniques of classical least squares (CLS), principal component regression (PCR), and partial least squares (PLS), as well as the univariate Beer’s law method have been described and compared, building students’ understanding by starting with the univariate method and progressing step by step into the multivariate methods. Equations for the production of regression vectors from training set spectral data is described and their use demonstrated for the prediction of constituent concentrations on a separate validation set of spectra. Extreme care is taken to ensure consistency in variable formatting of data matrices. This provides a key foundation to understanding how spectral data are manipulated using these different mathematical approaches for building quantitative regression models. Each method is applied to a real-world data set, and the results are discussed to show students the types of information that can be gleaned from each method. A training set comprised of 20 infrared absorbance spectra containing 3 constituents (benzene, polystyrene, and gasoline) of known composition are used to demonstrate the matrix operations for each regression method. A separate set of 12 real-world napalm samples (containing benzene, polystyrene and gasoline) are used as a validation set to demonstrate the ability to utilize the regression models on an unknown dataset. A toolbox (PNNL Chemometric Toolbox) written in MATLAB language is supplied in the Supplemental Information file and can be used as a companion for understanding the development and deployment of the chemometric algorithms described in this paper. The datasets of the infrared spectra are also supplied, allowing users to build and inspect the chemometric models on their own. Finally, the Toolbox includes scripts to assist users in loading their own datasets into MATLAB and performing CLS, PCR, and PLS on their data.

Upper-Division Undergraduate, Analytical Chemistry↗
DOE OSTI2024

SEAS Communication Engine: An Extensible, Flexible Wrapper for Co-Simulation Agents

When modeling and analyzing the power grid and other large scale systems, researchers often express scenarios as optimization problems and feed them into advanced software solvers. In order to allow multiple solvers to communicate with each other and share data from different domains, the National Renewable Energy Laboratory (NREL) and associated Department of Energy (DOE) labs have developed a software framework called the Hierarchical Engine for Large-scale Infrastructure Co-Simulation (HELICS). HELICS allows cosimulation via a collection of client libraries for different languages that can be called from the appropriate optimization software. However, these client libraries do not provide a higher level of abstraction beyond reading and writing data off of the shared HELICS bus. In this paper, we describe a new software library called the SEAS Communication Engine that exposes a higher-level API for running cosimulation problems. The SEAS Engine provides a class-based abstraction on top of the Python HELICS client, in order to allow users to implement their domain-specific cosimulations without needing to interact with core HELICS primitives. This will make adoption of HELICS and cosimulation in general easier, by exposing a simpler API. In the second part of the paper, we validate our library on a collection of different simulation examples, including the canonical IEEE 13 Bus Feeder. Lastly, we demonstrate using the SEAS Engine to directly call domain-specific code written in the Julia programming language. Our hope is that this will serve as a template for easily calling software in different programming languages via the SEAS Engine, thereby avoiding code duplication and complexity.

co-simulation↗
DOE OSTI2022

Fractional vortices, Z 2 gauge theory, and the confinement-deconfinement transition

Here, in this Letter we discuss the classical three-dimensional XY model whose nearest-neighbor interaction is a mixture of cos (θ i -θ j ) (ferromagnetic) and cos 2 (θ i -θ j ) (nematic). This model is dual to a theory with integer and half-integer vortices. While both types of vortices interact with a noncompact U(1) gauge field, the half-integer vortices interact with an extra interaction mediated by a Z 2 gauge field. We shall discuss the confinement-deconfinement transition of the half-integer vortices, the Wilson and the 't Hooft loops, and their mutual statistics in path integral language. In addition, we shall present a quantum version of the classical model which exhibits these physics.

75 CONDENSED MATTER PHYSICS, SUPERCONDUCTIVITY AND↗
DOE OSTI2022

Resilience Design Patterns: A Structured Approach to Resilience at Extreme Scale (V.2.0)

Reliability is a serious concern for future extreme-scale high-performance computing (HPC) systems. Projections based on the current generation of HPC systems and technology roadmaps suggest the prevalence of very high fault rates in future systems. The errors resulting from these faults will propagate and generate various kinds of failures, which may result in outcomes ranging from result corruptions to catastrophic application crashes. Therefore, the resilience challenge for extreme-scale HPC systems requires coordination between various hardware and software technologies that are capable of handling a broad set of fault models at accelerated fault rates. Also, due to practical limits on power consumption in future HPC systems, they are likely to embrace innovative architectures, increasing the levels of hardware and software complexities. Therefore, the techniques that seek to improve resilience must navigate the complex trade-off space between resilience and the overheads to power consumption and performance. While the HPC community has developed various resilience solutions, application-level techniques as well as system-based solutions, the solution space of HPC resilience techniques remains fragmented. There are no formal methods to integrate the various HPC resilience techniques into composite solutions, nor are there methods to holistically evaluate the adequacy and efficacy of such solutions in terms of their protection coverage, and their performance & power efficiency characteristics. Additionally, few implementations of current resilience solutions are portable to newer architectures and software environments that will be deployed on future systems. We developed a new structured approach to the management of HPC resilience using the concept of resilience-based design patterns. In general, a design pattern is a repeatable solution to a commonly occurring problem. We identified the well-known solutions that are commonly used to deal with faults, errors and failures in HPC systems. In the initial design patterns specification (version 1.0), we described the various solutions, which address specific problems in the design of resilient HPC environments, in the form of patterns. Each pattern describes a problem caused by a fault, error or failure event in an HPC environment, and then describes the core of the solution of the problem in such a way that this solution may be adapted to different systems and implemented at different layers of the system stack. The catalog of these resilience design patterns provides designers with a collection of design elements. To construct complete resilience solutions using combinations of various patterns, we defined a framework that enhances HPC designers' understanding of the important constraints and the opportunities for the design patterns to be implemented and deployed at various layers of the system stack. The design framework is also useful for establishing interfaces and mechanisms to coordinate flexible fault management across hardware and software components, as well as to consider the trade-off between performance, resilience, and power consumption when constructing a solution. The resilience design patterns specification version 1.1 included more detailed explanations of the pattern solutions, the context in which the patterns are applicable, and the implications for hardware or software design. It also provided several additional examples and detailed case studies to demonstrate the use of patterns to build realistic solutions. In version 1.2 of the specification document, we have improved the pattern descriptions, including graphical representations of the pattern components. These improvements are largely based on critical comments, feedback and suggestions received from pattern experts and readers of the previous versions of the specification. The pattern classification has been modified to further clarify the relationships between pattern categories. This version of the specification also introduces a pattern language for resilience design patterns. The pattern language presents the patterns in the catalog as a network, revealing the relations among the resilience patterns. The language provides designers with the means to explore alternative techniques for handling a specific fault model that may have different efficiency and complexity characteristics. Using the pattern language also enables the design and implementation of comprehensive resilience solutions as a set of interconnected resilience patterns that can be instantiated across layers of the system stack. The overall goal of this work is to provide hardware and software designers, as well as the users and operators of HPC systems, a systematic methodology for the design and evaluation of resilience technologies in HPC systems that keep scientific applications running to a correct solution in a timely and cost-efficient manner despite frequent faults, errors, and failures of various types. Version 2.0 expands the resilience design pattern classification and catalog to include self-stabilization patterns and reliability, availability and performance models for each structural pattern.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2020

High-Level Model Articulation with BuildingSync and OpenStudio

The use of the BuildingSync schema for describing the contents of buildings is becoming more common with its recent integration into the Audit Template tool as well as ASHRAE’s Building Energy Quotient (bEQ) web portal. Although BuildingSync was initially created to store and transfer data related to building energy audits (as defined by ASHRAE Standard 211), it has since been expanded to store the data needed to articulate fully defined physics-based building energy models. BuildingSync combined with abstracted high-level input methods defined in OpenStudio’s Standards project and the newly developed BuildingSync gem allows BuildingSync eXtensible Markup Language (XML) (Bray, Paoli, Sperberg-McQueen, Maler, Eve (Sun Microsystems, & Francois, 2008) document to be converted into OpenStudio models. Automatic generation of a building energy model from data collected during an audit 1) eliminates the need to separately generate an energy model, 2) improves consistency between model generation, and 3) simplifies evaluation of various energy efficiency measures. This manuscript will discuss this open-source project, including: development of the validation infrastructure necessary to provide formalized expectations of informational requirements for BuildingSync documents; development of the BuildingSync gem for translation of BuildingSync documents to OpenStudio models as well as example implementations; and finally, the manuscript will elaborate on the advantages and disadvantages of using high-level models generated from BuildingSync as surrogates to detailed models.

30 DIRECT ENERGY CONVERSION↗
DOE OSTI2026

Subseasonal Forecasting and MJO Teleconnections in Machine Learning Weather Prediction Models

Abstract In recent years, machine‐learning (ML) models trained on reanalysis data have rivaled physics‐based forecast models in terms of performance skill for global weather forecasting. With increased rollout stability, the question of how these models perform for subseasonal to seasonal (S2S, week 3–8) forecasting has emerged. In this study we run a large set of subseasonal hindcasts over 2004–2023 to evaluate two ML weather forecast models at the S2S time scale, SFNO‐HENS (Nvidia, fully ML) and NeuralGCM (Google Research, hybrid). Corresponding hindcasts from the European Centre for Medium‐Range Weather Forecasts (ECMWF) are used as a baseline for comparison to a physics‐based model. Because our focus is on predicting moisture transport over the Western United States between October and March, we evaluate the models' prediction skill for the Madden‐Julian Oscillation (MJO) and its associated teleconnections in the North Pacific. We find that both ML models are competitive with the ECWMF model, with comparable skill in predicting the North Pacific large‐scale circulation and the MJO at week 3 and beyond. Even though overall the mid‐latitude subseasonal prediction skill remains low, the ML models exhibit interesting behavior such as a realistic propagation of the MJO across the Maritime Continent and realistic teleconnections. A SFNO‐HENS sensitivity experiment with altered initial conditions in the tropics demonstrates the stability of the model, and it illustrates the capability of ML models to represent important physical processes of the atmosphere at the S2S time scale. Plain Language Summary Predicting weather patterns and precipitation a few weeks in advance (subseasonal time scale) is of great interest for stakeholders such as water managers in the Southwest United States (US), where arid conditions prevail. Subseasonal forecasts from traditional weather forecast models exhibit low skill in the region, limiting their applicability. Here we examine whether the recent breakthrough in weather forecasting made with machine learning/artificial intelligence models can translate to improved subseasonal forecasts. Recently‐developed machine learning models exhibit comparable skill to a state‐of‐the‐art physics‐based model for predicting weather patterns in the North Pacific/North America region, and associated moisture transport. The same applies to their skill in predicting the tropical pattern, the Madden‐Julian Oscillation, and its important remote perturbations over the midlatitude East Pacific and Southwest US. Additionally, a perturbation experiment carried out with one of the machine learning models illustrates their ability to not only predict the evolution of atmospheric fields, but also to learn and represent physical processes such as tropics‐extratropics Rossby wave propagation. Key Points Two machine learning weather forecast models exhibit state‐of‐the‐art prediction skill at the subseasonal time scale in the Pacific sector The models equal ECWMF in terms of Madden‐Julian oscillation (MJO) prediction skill, and they accurately predict the MJO propagation and associated teleconnections The two machine‐learning models represent key physical processes for subseasonal prediction, despite being trained for weather forecasting

Peings, Yannick↗