Search NASA⌕ Search

SEARCH · Search NASA

Results for “Error Metrics”

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 19 records

Effectiveness of various error metrics in SCEPTRE.

The objective of this project is to investigate accuracy of error metrics in SCEPTRE and produce useful benchmarks, identify metrics that do not work well, identify metrics that do work well, and produce easy to reference results.

99 GENERAL AND MISCELLANEOUS↗

WE-H-BRC-05: Catastrophic Error Metrics for Radiation Therapy

Purpose: Intuitive evaluation of complex radiotherapy treatments is impractical, while data transfer anomalies create the potential for catastrophic treatment delivery errors. Contrary to prevailing wisdom, logical scrutiny can be applied to patient-specific machine settings. Such tests can be automated, applied at the point of treatment delivery and can be dissociated from prior states of the treatment plan, potentially revealing errors introduced early in the process. Methods: Analytical metrics were formulated for conventional and intensity modulated RT (IMRT) treatments. These were designed to assess consistency between monitor unit settings, wedge values, prescription dose and leaf positioning (IMRT). Institutional metric averages for 218 clinical plans were stratified over multiple anatomical sites. Treatment delivery errors were simulated using a commercial treatment planning system and metric behavior assessed via receiver-operator-characteristic (ROC) analysis. A positive result was returned if the erred plan metric value exceeded a given number of standard deviations, e.g. 2. The finding was declared true positive if the dosimetric impact exceeded 25%. ROC curves were generated over a range of metric standard deviations. Results: Data for the conventional treatment metric indicated standard deviations of 3%, 12%, 11%, 8%, and 5 % for brain, pelvis, abdomen, lung and breast sites, respectively. Optimum error declaration thresholds yielded true positive rates (TPR) between 0.7 and 1, and false positive rates (FPR) between 0 and 0.2. Two proposed IMRT metrics possessed standard deviations of 23% and 37%. The superior metric returned TPR and FPR of 0.7 and 0.2, respectively, when both leaf position and MUs were modelled. Isolation to only leaf position errors yielded TPR and FPR values of 0.9 and 0.1. Conclusion: Logical tests can reveal treatment delivery errors and prevent large, catastrophic errors. Analytical metrics are able to identify errors in monitor units, wedging and leaf positions with favorable sensitivity and specificity. In part by Varian.

60 APPLIED LIFE SCIENCES↗

Error metrics for partially coherent wave fields

Lensless imaging methods that account for partial coherence have become very common in the past decade. However, there are no metrics in use for comparing partially coherent light fields, despite the widespread use of such metrics to compare fully coherent objects and wave fields. Here, we show how reformulating the mean squared error and Fourier ring correlation in terms of quantum state fidelity naturally generalizes them to partially coherent wave fields. We report these results fill an important gap in the lensless imaging literature and will enable quantitative assessments of the reliability and resolution of reconstructed partially coherent wave fields.

71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSIC↗

Accounting for Training Data Error in Machine Learning Applied to Earth Observations

Remote sensing, or Earth Observation (EO), is increasingly used to understand Earth system dynamics and create continuous and categorical maps of biophysical properties and land cover, especially based on recent advances in machine learning (ML). ML models typically require large, spatially explicit training datasets to make accurate predictions. Training data (TD) are typically generated by digitizing polygons on high spatial-resolution imagery, by collecting in situ data, or by using pre-existing datasets. TD are often assumed to accurately represent the truth, but in practice almost always have error, stemming from (1) sample design, and (2) sample collection errors. The latter is particularly relevant for image-interpreted TD, an increasingly commonly used method due to its practicality and the increasing training sample size requirements of modern ML algorithms. TD errors can cause substantial errors in the maps created using ML algorithms, which may impact map use and interpretation. Despite these potential errors and their real-world consequences for map-based decisions, TD error is often not accounted for or reported in EO research. Here we review the current practices for collecting and handling TD. We identify the sources of TD error, and illustrate their impacts using several case studies representing different EO applications (infrastructure mapping, global surface flux estimates, and agricultural monitoring), and provide guidelines for minimizing and accounting for TD errors. To harmonize terminology, we distinguish TD from three other classes of data that should be used to create and assess ML models: training reference data, used to assess the quality of TD during data generation; validation data, used to iteratively improve models; and map reference data, used only for final accuracy assessment. We focus primarily on TD, but our advice is generally applicable to all four classes, and we ground our review in established best practices for map accuracy assessment literature. EO researchers should start by determining the tolerable levels of map error and appropriate error metrics. Next, TD error should be minimized during sample design by choosing a representative spatio-temporal collection strategy, by using spatially and temporally relevant imagery and ancillary data sources during TD creation, and by selecting a set of legend definitions supported by the data. Furthermore, TD error can be minimized during the collection of individual samples by using consensus-based collection strategies, by directly comparing interpreted training observations against expert-generated training reference data to derive TD error metrics, and by providing image interpreters with thorough application-specific training. We strongly advise that TD error is incorporated in model outputs, either directly in bias and variance estimates or, at a minimum, by documenting the sources and implications of error. TD should be fully documented and made available via an open TD repository, allowing others to replicate and assess its use. To guide researchers in this process, we propose three tiers of TD error accounting standards. Finally, we advise researchers to clearly communicate the magnitude and impacts of TD error on map outputs, with specific consideration given to the likely map audience.

58 GEOSCIENCES↗

Evaluating the WFIP2 updates to the HRRR model using scanning Doppler lidar measurements in the complex terrain of the Columbia River Basin

The wind-energy (WE) industry relies on numerical weather prediction (NWP) forecast models as foundational or base models for many purposes, including wind-resource assessment and wind-power forecasting. During the Second Wind Forecast Improvement Project (WFIP2) in the Columbia River Basin of Oregon and Washington, a significant effort was made to improve NWP forecasts through focused model development, to include experimental refinements to the High Resolution Rapid Refresh (HRRR) model physics and horizontal grid spacing. In this study, the performance of an experimental version of HRRR that includes these refinements is tested against a control version, which corresponds to that of the operational HRRR run by National Oceanic and Atmospheric Administration/National Centers for Environmental Protection at the outset of WFIP2. Furthermore, the effects of horizontal grid resolution were also tested by comparing wind forecasts from the HRRR (with 3-km grid spacing) with those from a finer-resolution HRRR nest with 750-m grid spacing. Model forecasts are validated against accurate wind-profile measurements by three scanning, pulsed Doppler lidars at sites separated by a total distance of 71 km. Model skill and improvements in model skill, attributable to physics refinements and improved horizontal grid resolution, varied by season, by site, and during periods of atmospheric phenomena relevant to WE. In general, model errors were the largest below 150 m above ground level (AGL). Experimental HRRR refinements tended to reduce the mean absolute error (MAE) and other error metrics for many conditions, but degradation in skill (increased MAE) was noted below 150 m AGL at the two lowest-elevation sites at night. Finer resolution was found to produce the most significant reductions in the error metrics.

17 WIND ENERGY↗

Short-term electricity load forecasting: Application-driven evaluation of machine learning models across spatial and temporal scales

As we transition towards a decarbonized economy, the integration of variable renewable energy resources and new demands (e.g., electric vehicles, heat pumps) into the electricity grid places unprecedented pressure on grid operators to effectively anticipate and manage peak load. In this context, machine learning algorithms are proving to be indispensable for accurate short-term load forecasting, a crucial task to address these challenges. This study benchmarks 6 machine learning algorithms, including three neural networks and three tree-based algorithms, across various levels of spatial aggregation and time horizons (1, 4, 8, 24, and 48 h). The central contribution of this work is the comparison and analysis of load forecasting models not only based on statistical metrics, but also based on a novel error metric, which evaluates the cost implications of forecast errors for power system stakeholders. Results show that tree-based models outperform neural networks, based on statistical metrics, and yield less skewed error distributions for most spatial scales. However, through the lens of the novel error metric, neural networks are the more competitive choice, especially for forecast horizons that exceed 8 h. The study concludes with actionable recommendations to grid operators and highlights the need for the development of error metrics that link forecasting accuracy to operational costs. To promote transparency and open science, the datasets and Python code are open-sourced via a supplementary repository.

Houben, Nikolaus↗

Comparing field data using Alpert multi-wavelets

In this paper we introduce a method to compare sets of full-field data using Alpert tree-wavelet transforms. The Alpert tree-wavelet methods transform the data into a spectral space allowing the comparison of all points in the fields by comparing spectral amplitudes. The methods are insensitive to translation, scale and discretization and can be applied to arbitrary geometries. This makes them especially well suited for comparison of field data sets coming from two different sources such as when comparing simulation field data to experimental field data. We have developed both global and local error metrics to quantify the error between two fields. Additionally, we verify the methods on two-dimensional and three-dimensional discretizations of analytical functions. Furthermore, we then deploy the methods to compare full-field strain data from a simulation of elastomeric syntactic foam.

42 ENGINEERING↗

Physics-informed transformation toward improving the machine-learned NLTE models of ICF simulations

The integration of machine-learning techniques into inertial confinement fusion (ICF) simulations has emerged as a powerful approach for enhancing computational efficiency. By replacing the costly nonlocal thermodynamic equilibrium (NLTE) model with machine-learning models, significant reductions in calculation time have been achieved. However, determining how to optimize machine-learning-based NLTE models in order to match ICF simulation dynamics remains challenging, underscoring the need for physically relevant error metrics and strategies to enhance model accuracy with respect to these metrics. Thus, we propose novel physics-informed transformations designed to emphasize energy transport, use these transformations to establish new error metrics, and demonstrate that they yield smaller errors within reduced principal-component spaces compared to conventional transformations. Published by the American Physical Society 2025

70 PLASMA PHYSICS AND FUSION TECHNOLOGY↗

Intercomparison of wind and wave data from the ECMWF Reanalysis Interim and the NCEP Climate Forecast System Reanalysis

The recent release of the ECMWF Reanalysis Interim (ERA-I) and NCEP Climate Forecast System Reanalysis (CFSR) allows for studies of global climate and its cycles with unprecedented detail. While the developers have performed verification and validation, there is little information on their relative performance in particular related to their use in ocean modeling. This study focuses on the intercomparison of wind and wave reanalysis datasets from ERA-I and CFSR utilizing the same set of altimetry and buoy observations and error metrics to assess their consistency in time and space. Both products have good spatial homogeneity with consistent levels of errors in the Northern and Southern Hemispheres. ERA-I proves to be homogenous through time, while CFSR exhibits an abrupt decrease in the level of errors in the Southern Ocean beginning 1994. ERA-I generally underestimates the wind speed and wave height with lower standard deviations in comparison to observations, but maintains slightly better error metrics. Despite having a slightly positive bias, CFSR provides a better description of the variability of the observations and improved performance in the upper percentiles associated with extreme events. Altogether ERA-I has better homogeneity through time deeming it more reliable for modeling of long-term processes; however caution must be applied with analysis of the upper percentiles.

16 TIDAL AND WAVE POWER↗

MS25: Materials Science-Focused Benchmark Data Set for Machine Learning Interatomic Potentials

Here, we present MS25, a benchmark data set for evaluating machine learning interatomic potentials (MLIPs) across diverse materials-relevant systems including MgO surfaces, liquid water, zeolites, a catalytic Pt surface reaction, high-entropy alloys (HEAs), and disordered Zr-oxides. Five MLIP architectures (MACE, NequIP, Allegro, MTP, and Torch-ANI) are trained and tested, focusing not only on traditional metrics (energies, forces, and stresses) but also explicitly validating derived physical observables such as lattice constants, volumes, and reaction barriers. We find that most models reach comparable accuracy on standard error metrics across the simple systems, although equivariant MLIPs offer 1.5–2× improvements over nonequivariant MLIPs in energy and force error for structurally complex or compositionally disordered environments such as HEAs and Zr–O systems. Our analysis highlights that low errors in energy and force predictions do not guarantee reliable observables, emphasizing the necessity of explicit validation. We demonstrate limitations in cross-framework transferability, as models trained on one zeolite framework (CHA) fail to reliably generalize to predictions of structurally distinct frameworks (e.g., MFI). Size-extensive tests show some dependence on system size for MgO, resulting from forced periodicity. The HEA and Zr–O data sets are identified as challenging tests for future benchmarks and MLIP model architecture developments as they show significant differentiation in error between MLIP architectures and are still relatively difficult at 1000 training images. Moving forward, we recommend that benchmarking efforts shift their focus from marginal accuracy improvements in energy and force errors toward identifying and understanding model failure modes, rigorously assessing transferability, and evaluating how their errors affect observable predictions. For researchers looking to choose an MLIP architecture, we suggest selecting equivariant MLIP architectures if the complexity of the system is a challenge. For simple materials problems, auxiliary features such as integration with molecular dynamics engines, trade-offs between computational data set generation cost vs MLIP inference speed, and framework integration may play a more important decision factor than small differences in error metrics that are unlikely to matter for production-level research.

chemical structure↗

Validating automated resonance evaluation with synthetic data

The integrity and precision of nuclear data are crucial for a broad spectrum of applications, from national security and nuclear reactor design to medical diagnostics, where the associated uncertainties can significantly impact outcomes. A substantial portion of uncertainty in nuclear data originates from the subjective biases in the evaluation process, a crucial phase in the nuclear data production pipeline. Recent advancements indicate that automation of certain routines can mitigate these biases, thereby standardizing the evaluation process and enhancing reproducibility. This research aims to provide a methodology, framework, and metrics for the validation of automated nuclear data evaluation software leveraging high-quality synthetic data that closely mimic real experimental observables. An introduced error metric provides a scale and intuitive measure of the evaluation quality by quantifying the estimate’s accuracy and performance across the specified energy range. Synthetic data provides access to experimental observables and underlying resonance parameters, enabling comparison of different evaluations. The methodology is demonstrated using Ta-181 isotope data in the resolved resonance region. The Automated Resonance Identification Subroutine (ARIS), which operates without prior resonance information, was used to test and showcase the framework’s capabilities utilizing the proposed error metrics. The results demonstrate the effectiveness of the proposed approach and framework for optimizing software parameters and testing hypotheses through “what-if” controlled experiments, such as modifying assumptions about experimental conditions or average resonance parameters.

22 GENERAL STUDIES OF NUCLEAR REACTORS↗

Development of a Method for Shape Optimization for a Gas Turbine Fuel Injector Design Using Metal-Additive Manufacturing

Adjoint shape optimization has enabled physics-based optimal designs for aerodynamic surfaces. Additive manufacturing (AM) makes it possible to manufacture complex shapes. However, there has been a gap between optimal and manufacturable surfaces due to the inherent limitations of commercial computational fluid dynamics (CFD) codes to implement geometric constraints during adjoint computation. In such cases, the design sensitivities are exported and used to perform constrained shape modifications using parametric information stored in computer aided design (CAD) files to satisfy manufacturability constraints. However, modifying the design using adjoint methods in CFD solvers and performing constrained shape modification in CAD can lead to inconsistencies due to different shape parameterization schemes. This paper describes a method to enable the simultaneous optimization of the fluid domain and impose AM manufacturability constraints, resolving one of the key issues of geometry definition for isogeometric analysis. Similar to a grid convergence study, the proposed method verifies the consistencies between shape parameterization techniques present within commercial CAD and CFD software during mesh movement as a part of the adjoint shape optimization routine. By identifying the appropriate parameters essential to a shape optimization study, the error metric between the different parameterization techniques converges to demonstrate sufficient consistencies for justifiable exchange of data between CAD and CFD. For the identified shape optimization parameters, the error metric to measure the deviation between the two parameterization schemes lies within the AM laser-powder bed fusion (L-PBF) process tolerance. Additionally, comparison for subsequent objective function calculations between iterations of the optimization loop showed acceptable differences within 1% variation between the modified geometries obtained using the two parameterization schemes. This method provides justification for the use of multiphysics guided adjoint design sensitivities computed in CFD software to perform shape modifications in CAD to incorporate AM manufacturability constraints during the shape optimization loop such that optimal designs are also additively manufacturable.

33 ADVANCED PROPULSION SYSTEMS↗

Validation of wind resource and energy production simulations for small wind turbines in the United States

Abstract. Due to financial and temporal limitations, the small wind community relies upon simplified wind speed models and energy production simulation tools to assess site suitability and produce energy generation expectations. While efficient and user-friendly, these models and tools are subject to errors that have been insufficiently quantified at small wind turbine heights. This study leverages observations from meteorological towers and sodars across the United States to validate wind speed estimates from the Wind Integration National Dataset (WIND) Toolkit, the European Centre for Medium-Range Weather Forecasts (ECMWF) Reanalysis v5 (ERA5), and the Modern-Era Retrospective Analysis for Research and Applications, version 2 (MERRA-2), revealing average biases within ±0.5 m s−1 at small wind hub heights. Observations from small wind turbines across the United States provide references for validating energy production estimates from the System Advisor Model (SAM), Wind Report, MyWindTurbine.com, and Global Wind Atlas 3 (GWA3), which are seen to overestimate actual annual capacity factors by 2.5, 4.2, 11.5, and 7.3 percentage points, respectively. In addition to quantifying the error metrics, this paper identifies sources of model and tool discrepancies, noting that interannual fluctuation in the wind resource, wind speed class, and loss assumptions produces more variability in estimates than different horizontal and vertical interpolation techniques. The results of this study provide small wind installers and owners with information about these challenges to consider when making performance estimates and thus possible adjustments accordingly. Looking to the future, recognizing these error metrics and sources of discrepancies provides model and tool researchers and developers with opportunities for product improvement that could positively impact small wind customer confidence and the ability to finance small wind projects.

17 WIND ENERGY↗

Modeling high-efficiency extreme ultraviolet etched multilayer phase-shift masks

Achieving high-throughput extreme ultraviolet (EUV) patterning remains a major challenge due to low source power; phase-shift masks can help solve this challenge for dense features near the resolution limit by creating brighter images than traditional absorber masks when illuminated with the same source power. We explore applications of etched multilayer phase-shift masks for EUV lithography, both in the current-generation 0.33 NA and next-generation 0.55 NA systems. We derive analytic formulas for the thin-mask throughput gains, which are 2.42× for lines and spaces and 5.86× for contacts compared with an absorber mask with dipole and quadrupole illumination, respectively. Using rigorous finite-difference time-domain simulations, we quantify variations in these gains by pitch and orientation, finding 87% to 113% of the thin-mask value for lines and spaces and a 91% to 99% for contacts. We introduce an edge placement error metric, which accounts for CD errors, relative feature motion, and telecentricity errors, and use this metric both to optimize mask designs for individual features and to explore which features can be printed on the same mask. Furthermore, we find that although partial coherence shrinks the process window, at an achievable sigma of 0.2 we obtain a depth of focus of 340 nm and an exposure latitude of 39.2%, suggesting that partial coherence will not limit the feasibility of this technology. Finally, we show that many problems such as sensitivity to etch uniformity can be greatly mitigated using a central obscuration in the imaging pupil.

42 ENGINEERING↗

High-throughput cancer hypothesis testing with an integrated PhysiCell-EMEWS workflow

BackgroundCancer is a complex, multiscale dynamical system, with interactions between tumor cells and non-cancerous host systems. Therapies act on this combined cancer-host system, sometimes with unexpected results. Systematic investigation of mechanistic computational models can augment traditional laboratory and clinical studies, helping identify the factors driving a treatment's success or failure. However, given the uncertainties regarding the underlying biology, these multiscale computational models can take many potential forms, in addition to encompassing high-dimensional parameter spaces. Therefore, the exploration of these models is computationally challenging. We propose that integrating two existing technologiesone to aid the construction of multiscale agent-based models, the other developed to enhance model exploration and optimizationcan provide a computational means for high-throughput hypothesis testing, and eventually, optimization.ResultsIn this paper, we introduce a high throughput computing (HTC) framework that integrates a mechanistic 3-D multicellular simulator (PhysiCell) with an extreme-scale model exploration platform (EMEWS) to investigate high-dimensional parameter spaces. We show early results in applying PhysiCell-EMEWS to 3-D cancer immunotherapy and show insights on therapeutic failure. We describe a generalized PhysiCell-EMEWS workflow for high-throughput cancer hypothesis testing, where hundreds or thousands of mechanistic simulations are compared against data-driven error metrics to perform hypothesis optimization.ConclusionsWhile key notational and computational challenges remain, mechanistic agent-based models and high-throughput model exploration environments can be combined to systematically and rapidly explore key problems in cancer. These high-throughput computational experiments can improve our understanding of the underlying biology, drive future experiments, and ultimately inform clinical practice.

Agent-based model↗

Further Investigation of Employing Binomial Small-Sample Estimator in M-LROM

Reduced Order Modeling (ROM) is essential to enable the identification of the so-called active parameter subspace for high-fidelity models that are employed in computationally intensive analyses such as uncertainty characterization and data assimilation. Given the complexity of models for real-world application, brute force methods to identify the active subspace may not be computationally efficient to realize the value of reduction, since in realistic problems, the full scale model must be executed few hundred times, which may be computationally taxing. Earlier work has introduced a new approach, which represents a further development of our previous work on ROM, referred to as multi-level ROM (M-LROM). In M-LROM, approximate models are employed to identify the active subspace instead of the full scale model. In this summary, we present our recent results on M-LROM, specifically in regard to the construction of reliable error metrics that can bound the errors resulting from the reduction. The idea is to employ a lower fidelity model, i.e., an approximation of the full scale or high fidelity model, to build the active subspace, and via the proposed error bound, the quality of the active subspace can be determined by calculating an upper bound on the errors resulting from the reduction. For this approach to be practical, the error bound must be realistic, implying that it should be close to the actual errors expected, and not much higher. Therefore, an important goal of this summary is to compare different error bounds in terms of the relation between their estimated bounds to the actual errors. In particular, we compare the proposed error bound which numerically construct an estimator using binomial distribution to the widely used small-sample norm estimators. Unlike most probabilistic error estimation methods, the proposed bound employs a binomial distribution to construct the estimator whereas in other conventional methods, the distribution employed is typically the standard Gaussian, normalized over a unit hypersphere as initially proposed by the seminal paper of J. Dixon on random matrix theory. Previous publications provided evidence that the binomial distribution calculates more realistic bounds as compared to the Gaussian distribution, which tends to be conservative, implying that the bound is much bigger than the actual errors, which diminishes the value of the reduction, as now must increase the size of the active subspace to reach the desired upper bound on the reduction errors. This work has provided a rigorous approach by which error bounds on the reduction errors can be established without having to execute the high fidelity model during the construction of the active subspace. The high fidelity model is only need to test the error bound using small number of executions. This methodology paves the way for a more efficient application of ROM to complex engineering models, where standard ROM techniques may be too taxing to optimally realize their value in practical situations. (authors)

22 GENERAL STUDIES OF NUCLEAR REACTORS↗

Improved Representation of Horizontal Variability and Turbulence in Mesoscale Simulations of an Extended Cold-Air Pool Event

Abstract Cold-air pools (CAPs), or stable atmospheric boundary layers that form within topographic basins, are associated with poor air quality, hazardous weather, and low wind energy output. Accurate prediction of CAP dynamics presents a challenge for mesoscale forecast models in part because CAPs occur in regions of complex terrain, where traditional turbulence parameterizations may not be appropriate. This study examines the effects of the planetary boundary layer (PBL) scheme and horizontal diffusion treatment on CAP prediction in the Weather Research and Forecasting (WRF) Model. Model runs with a one-dimensional (1D) PBL scheme and Smagorinsky-like horizontal diffusion are compared with runs that use a new three-dimensional (3D) PBL scheme to calculate turbulent fluxes. Simulations are completed in a nested configuration with 3-km/750-m horizontal grid spacing over a 10-day case study in the Columbia River basin, and results are compared with observations from the Second Wind Forecast Improvement Project. Using event-averaged error metrics, potential temperature and wind speed errors are shown to decrease both with increased horizontal grid resolution and with improved treatment of horizontal diffusion over steep terrain. The 3D PBL scheme further reduces errors relative to a standard 1D PBL approach. Error reduction is accentuated during CAP erosion, when turbulent mixing plays a more dominant role in the dynamics. Last, the 3D PBL scheme is shown to reduce near-surface overestimates of turbulence kinetic energy during the CAP event. The sensitivity of turbulence predictions to the master length-scale formulation in the 3D PBL parameterization is also explored. Significance Statement In this article, we demonstrate how a new framework for modeling atmospheric turbulence improves cold pool predictions, using a case study from January 2017 in the Columbia River basin (U.S. Pacific Northwest). Cold pools are regions of cold, stagnant air that form within valleys or basins, and improved forecasts could help to mitigate the risks they pose to air quality, transportation, and wind energy production. For the chosen case study, our tests show a reduction in temperature and wind speed errors by up to a factor of 2–3 relative to standard model options. These results strongly motivate continued development of the framework as well as its application to other complex weather events.

17 WIND ENERGY↗