Search NASA⌕ Search

SEARCH · Search NASA

Results for “variational methods”

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 1,693 records · Page 94

Anatomy structure creation and editing using 3D implicit surfaces

Purpose: To accurately reconstruct, and interactively reshape 3D anatomy structures' surfaces using small numbers of 2D contours drawn in the most visually informative views of 3D imagery. The innovation of this program is that the number of 2D contours can be very much smaller than the number of transverse sections, even for anatomy structures spanning many sections. This program can edit 3D structures from prior segmentations, including those from autosegmentation programs. The reconstruction and surface editing works with any image modality. Methods: Structures are represented by variational implicit surfaces defined by weighted sums of radial basis functions (RBFs). Such surfaces are smooth, continuous, and closed and can be reconstructed with RBFs optimally located to efficiently capture shape in any combination of transverse (T), sagittal (S), and coronal (C) views. The accuracy of implicit surface reconstructions was measured by comparisons with the corresponding expert-contoured surfaces in 103 prostate cancer radiotherapy plans. Editing a pre-existing surface is done by overdrawing its profiles in image views spanning the affected part of the structure, deleting an appropriate set of prior RBFs, and merging the remainder with the new edit contour RBFs. Two methods were devised to identify RBFs to be deleted based only on the geometry of the initial surface and the locations of the new RBFs. Results: Expert-contoured surfaces were compared with implicit surfaces reconstructed from them over varying numbers and combinations of T/S/C planes. Studies revealed that surface-surface agreement increases monotonically with increasing RBF-sample density, and that the rate of increase declines over the same range. These trends were observed for all surface agreement metrics and for all the organs studied--prostate, bladder, and rectum. In addition, S and C contours may convey more shape information than T views for CT studies in which the axial slice thickness is greater than the pixel size. Surface editing accuracy likewise improves with larger sampling densities, and the rate of improvement similarly declines over the same conditions. Conclusions: Implicit surfaces based on RBFs are accurate representations of anatomic structures and can be interactively generated or modified to correct segmentation errors. The number of input contours is typically smaller than the number of T contours spanned by the structure.

62 RADIOLOGY AND NUCLEAR MEDICINE↗

Hints of correlation between broad-line and radio variations for 3C 120

In this paper, we investigate the correlation between broad-line and radio variations for the broad-line radio galaxy 3C 120. By the z-transformed discrete correlation function method and the model-independent flux randomization/random subset selection (FR/RSS) Monte Carlo method, we find that broad Hβ line variations lead the 15 GHz variations. The FR/RSS method shows that the Hβ line variations lead the radio variations by a factor of τ{sub ob} = 0.34 ± 0.01 yr. This time lag can be used to locate the position of the emitting region of radio outbursts in the jet, on the order of ∼5 lt-yr from the central engine. This distance is much larger than the size of the broad-line region. The large separation of the radio outburst emitting region from the broad-line region will observably influence the gamma-ray emission in 3C 120.

79 ASTRONOMY AND ASTROPHYSICS↗

Reduction of ring artifacts in CBCT: Detection and correction of pixel gain variations in flat panel detectors

Purpose: In using flat panel detectors (FPD) for cone beam computed tomography (CBCT), pixel gain variations may lead to structured nonuniformities in projections and ring artifacts in CBCT images. Such gain variations can be caused by change in detector entrance exposure levels or beam hardening, and they are not accounted by conventional flat field correction methods. In this work, the authors presented a method to identify isolated pixel clusters that exhibit gain variations and proposed a pixel gain correction (PGC) method to suppress both beam hardening and exposure level dependent gain variations. Methods: To modulate both beam spectrum and entrance exposure, flood field FPD projections were acquired using beam filters with varying thicknesses. “Ideal” pixel values were estimated by performing polynomial fits in both raw and flat field corrected projections. Residuals were calculated by taking the difference between measured and ideal pixel values to identify clustered image and FPD artifacts in flat field corrected and raw images, respectively. To correct clustered image artifacts, the ratio of ideal to measured pixel values in filtered images were utilized as pixel-specific gain correction factors, referred as PGC method, and they were tabulated as a function of pixel value in a look-up table. Results: 0.035% of detector pixels lead to clustered image artifacts in flat field corrected projections, where 80% of these pixels were traced back and linked to artifacts in the FPD. The performance of PGC method was tested in variety of imaging conditions and phantoms. The PGC method reduced clustered image artifacts and fixed pattern noise in projections, and ring artifacts in CBCT images. Conclusions: Clustered projection image artifacts that lead to ring artifacts in CBCT can be better identified with our artifact detection approach. When compared to the conventional flat field correction method, the proposed PGC method enables characterization of nonlinear pixel gain variations as a function of change in x-ray spectrum and intensity. Hence, it can better suppress image artifacts due to beam hardening as well as artifacts that arise from detector entrance exposure variation.

62 RADIOLOGY AND NUCLEAR MEDICINE↗

On the determination of reference levels for quality assurance of flattening filter free photon beams in radiation therapy

Purpose: New definitions for some dosimetric parameters for use in quality assurance of flattening filter free (FFF) beams generated by medical linear accelerators have been suggested. The present study aims to validate these suggestions and to propose possible reference levels. Methods: The main characteristics of FFF photon beams were described in terms of: field size, penumbra, unflatness, slope, and peak-position parameters. Data were collected for 6 and 10 MV-FFF beams from three different Varian TrueBeam Linacs. Measurements were performed with a 2D-array (Starcheck system from PTW-Freiburg) and with the portal dosimetry method GLAaS utilizing the build-in portal imager of TrueBeam. Data were also compared to ion chamber measurements. A cross check validation has been performed on a FFF beam of 6 MV generated by a Varian Clinac-iX upgraded to FFF capability. Results : All the parameters suggested to characterize the FFF beams resulted easily measurable and little variation was observed among different Linacs. Referring to two reference field sizes of 10 × 10 and 20 × 20 cm{sup 2}, at SDD = 100 cm and d = dmax, from the portal dosimetry data, the following results (averaging X and Y profiles) were obtained. Field size: 9.95 ± 0.02 and 19.98 ± 0.03 cm for 6 MV-FFF (9.94 ± 0.02 and 19.98 ± 0.03 cm for 10 MV-FFF). Penumbra: 2.7 ± 0.3 and 2.9 ± 0.3 mm for 6 MV-FFF (3.1 ± 0.2 and 3.3 ± 0.3 for 10 MV-FFF). Unflatness: 1.11 ± 0.01 and 1.25 ± 0.01 for 6 MV-FFF (1.21 ± 0.01 and 1.50 ± 0.01 for 10 MV-FFF). Slope: 0.320 ± 0.020%/mm and 0.43 ± 0.015%/mm for 6 MV-FFF (0.657 ± 0.023%/mm and 0.795 ± 0.017%/mm for 10 MV-FFF). Peak Position −0.2 ± 0.2 and −0.4 ± 0.2 mm for 6 MV-FFF (−0.3 ± 0.2 and 0.7 ± 0.3 mm for 10 MV-FFF). Results would depend upon measurement depth. With thresholds set to at least 95% confidence level from the measured data and to account for possible variations between detectors and methods and experimental settings, a tolerance set of: 1 mm for field size and penumbra, 0.04 for unflatness, 0.1%/mm for slope, and 1 mm for peak position could be proposed from our data. Conclusions : The parameters proposed for the characterization and routine control of stability of profiles of FFF beams appear to be a viable solution with a strong similarity to the conventional parameters used for flattened beams. The results from three different TrueBeams and the cross-validation against a Clinac-iX suggested the possible generalization of the methods and the possibility to use common tolerances for the parameters. The data showed also the reproducibility of beam characteristics among different systems (of the same vendor) and the resulting parameter values could therefore be possibly generalized.

62 RADIOLOGY AND NUCLEAR MEDICINE↗

Response of Water Use Efficiency to Global Environmental Change Based on Output From Terrestrial Biosphere Models

Here, water use efficiency (WUE), defined as the ratio of gross primary productivity and evapotranspiration at the ecosystem scale, is a critical variable linking the carbon and water cycles. Incorporating a dependency on vapor pressure deficit, apparent underlying WUE (uWUE) provides a better indicator of how terrestrial ecosystems respond to environmental changes than other WUE formulations. Here we used 20th century simulations from four terrestrial biosphere models to develop a novel variance decomposition method. With this method, we attributed variations in apparent uWUE to both the trend and interannual variation of environmental drivers. The secular increase in atmospheric CO 2 explained a clear majority of total variation (66 ± 32%: mean ± one standard deviation), followed by positive trends in nitrogen deposition and climate, as well as a negative trend in land use change. In contrast, interannual variation was mostly driven by interannual climate variability. To analyze the mechanism of the CO 2 effect, we partitioned the apparent uWUE into the transpiration ratio (transpiration over evapotranspiration) and potential uWUE. The relative increase in potential uWUE parallels that of CO 2 , but this direct CO 2 effect was offset by 20 ± 4% by changes in ecosystem structure, that is, leaf area index for different vegetation types. However, the decrease in transpiration due to stomatal closure with rising CO 2 was reduced by 84% by an increase in leaf area index, resulting in small changes in the transpiration ratio. CO 2 concentration thus plays a dominant role in driving apparent uWUE variations over time, but its role differs for the two constituent components: potential uWUE and transpiration.

54 ENVIRONMENTAL SCIENCES↗

Characteristics of the Magnetic Resonance WPT System Applying a Frequency Converter

Recently, studies on the commercialization of magnetic resonance wireless power transfer (WPT) systems are underway domestically and internationally. The magnetic resonance WPT technology enables power transmission between transmitter and receiver coils with the same resonance frequency. If the resonance frequencies of the transmitter coil and receiver coil are different, the efficiency of the WPT will rapidly decrease. Various organizations, such as A4WP, have been striving to standardize resonance frequencies, but the countries around the world are sticking to their own distinctive standards based on the frequency bands available in their territories. Therefore, each country uses a slightly different resonance frequency band. In this paper, a frequency converter that enables WPT when the transmitter and receiver coils have different resonance frequencies is proposed. The frequency converter consists of two coils, and its structure is very simple. The two coils transmit power using the magnetic induction method, and frequency variation is possible using variable capacitors. It was confirmed from the experiment that WPT is possible even if the resonance frequencies of the transmitter and receiver coils are different.

75 CONDENSED MATTER PHYSICS, SUPERCONDUCTIVITY AND↗

Bayesian inference of heterogeneous epidemic models: Application to COVID-19 spread accounting for long-term care facilities

Here we propose a high dimensional Bayesian inference framework for learning heterogeneous dynamics of a COVID-19 model, with a specific application to the dynamics and severity of COVID-19 inside and outside long-term care (LTC) facilities. We develop a heterogeneous compartmental model that accounts for the heterogeneity of the time-varying spread and severity of COVID-19 inside and outside LTC facilities, which is characterized by time-dependent stochastic processes and time-independent parameters in ~ 1500 dimensions after discretization. To infer these parameters, we use reported data on the number of confirmed, hospitalized, and deceased cases with suitable post-processing in both a deterministic inversion approach with appropriate regularization as a first step, followed by Bayesian inversion with proper prior distributions. To address the curse of dimensionality and the ill-posedness of the high-dimensional inference problem, we propose use of a dimension-independent projected Stein variational gradient descent method, and demonstrate the intrinsic low-dimensionality of the inverse problem. We present inference results with quantified uncertainties for both New Jersey and Texas, which experienced different epidemic phases and patterns. Moreover, we also present forecasting and validation results based on the empirical posterior samples of our inference for the future trajectory of COVID-19.

42 ENGINEERING↗

CO 2 induced changes in Mount Simon sandstone: Understanding links to post CO 2 injection monitoring, seismicity, and reservoir integrity

The purpose of this study is to quantify geochemical reactions of CO 2 and brine with subsurface samples taken from the Mt. Simon sandstone and identify any potential alterations of the geomechanical rock properties that could lead to changes observable in seismic monitoring or result in changes of micro seismicity such as those observed at the Illinois Basin-Decatur Project (IBDP) site. Two Mt. Simon sandstone samples from 6919.3 feet (2109.0 m) and 6926.1 feet (2111.1 m) depth were exposed to supercritical CO 2 (scCO 2 ) dissolved in brine at in-situ reservoir conditions for one month. Geochemical, spectral, scanning electron microscopy, and petrophysical methods were used to analyze the samples before and after the one-month exposure. Significant changes were observed. Multiple geomechanical properties were chosen to form the framework against which to interrogate the petrophysical data: Young’s modulus (E), Poisson’s ratio (ν), lambda·rho (λρ), and mu·rho (μρ). In this study we conclude that framework composition, porosity, heterogeneities, effective pressure, and reactive geochemistry are first order controls on trends in the E-μ and λρ-μρ cross plot spaces. Changes in porosity, permeability, dynamic moduli, and brittleness with exposure to these fluids were observed. No change in ultrasonic P-wave attenuation (Q P ) was observed. Geochemical alteration causes a distinct shift in λρ-μρ in both samples as well as changes in E, ν, and P and S wave seismic velocity values. Furthermore, these observations could provide insight into subsurface monitoring using seismic methods including amplitude variation with offset (AVO) classification.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗

WENO schemes on arbitrary unstructured meshes for laminar, transitional and turbulent flows

This paper presents the development and implementation of weighted-essentially-non-oscillatory (WENO) schemes for viscous flows on arbitrary unstructured grids. WENO schemes up to fifth-order accurate have been implemented in conjunction with hybrid and non-hybrid unstructured grids. The schemes are investigated with reference to numerical and experimental results for the Taylor–Green vortex, as well as for laminar and turbulent flows around a sphere, and the turbulent shock-wave boundary layer interaction flow problem. The results show that the accuracy of the schemes depends on the arbitrariness of shape and orientation of the unstructured mesh elements, as well as the compactness of directional stencils. The WENO schemes provide a more accurate numerical framework compared to second-order and third-order total variation diminishing (TVD) methods, however, the fifth-order version of the schemes is computationally too expensive to make the schemes practically usable. On the other hand, the third-order variant offers an excellent numerical framework in terms of accuracy and computational cost compared to the fifth-order WENO and second-order TVD schemes. Parallelisation of the CFD code (henceforth labelled as UCNS3D), where the schemes have been implemented, shows that the present methods offer very good scalable performance.

42 ENGINEERING↗

A scalable geometric multigrid solver for nonsymmetric elliptic systems with application to variable-density flows

A geometric multigrid algorithm is introduced for solving nonsymmetric linear systems resulting from the discretization of the variable density Navier–Stokes equations on nonuniform structured rectilinear grids and high-Reynolds number flows. The restriction operation is defined such that the resulting system on the coarser grids is symmetric, thereby allowing for the use of efficient smoother algorithms. To achieve an optimal rate of convergence, the sequence of interpolation and restriction operations are determined through a dynamic procedure. A parallel partitioning strategy is introduced to minimize communication while maintaining the load balance between all processors. To test the proposed algorithm, we consider two cases: 1) homogeneous isotropic turbulence discretized on uniform grids and 2) turbulent duct flow discretized on stretched grids. Testing the algorithm on systems with up to a billion unknowns shows that the cost varies linearly with the number of unknowns. This O (N) behavior confirms the robustness of the proposed multigrid method regarding ill-conditioning of large systems characteristic of multiscale high-Reynolds number turbulent flows. The robustness of our method to density variations is established by considering cases where density varies sharply in space by a factor of up to 104, showing its applicability to two-phase flow problems. Strong and weak scalability studies are carried out, employing up to 30,000 processors, to examine the parallel performance of our implementation. Excellent scalability of our solver is shown for a granularity as low as 104 to 105 unknowns per processor. At its tested peak throughput, it solves approximately 4 billion unknowns per second employing over 16,000 processors with a parallel efficiency higher than 50%.

Computer Science↗

A Kaczmarz-inspired approach to accelerate the optimization of neural network wavefunctions

Neural network wavefunctions optimized using the variational Monte Carlo method have been shown to produce highly accurate results for the electronic structure of atoms and small molecules, but the high cost of optimizing such wavefunctions prevents their application to larger systems. We propose the Subsampled Projected-Increment Natural Gradient Descent (SPRING) optimizer to reduce this bottleneck. SPRING combines ideas from the recently introduced minimum-step stochastic reconfiguration optimizer (MinSR) and the classical randomized Kaczmarz method for solving linear least-squares problems. We demonstrate that SPRING outperforms both MinSR and the popular Kronecker-Factored Approximate Curvature method (KFAC) across a number of small atoms and molecules, given that the learning rates of all methods are optimally tuned. For example, on the oxygen atom, SPRING attains chemical accuracy after forty thousand training iterations, whereas both MinSR and KFAC fail to do so even after one hundred thousand iterations.

97 MATHEMATICS AND COMPUTING↗

Clinical practice and evaluation of electronic portal imaging device for VMAT quality assurance

Volumetric-modulated arc therapy (VMAT) is a novel extension of the intensity-modulated radiation therapy (IMRT) technique, which has brought challenges to dose verification. To perform VMAT pretreatment quality assurance, an electronic portal imaging device (EPID) can be applied. This study's aim was to evaluate EPID performance for VMAT dose verification. First, dosimetric characteristics of EPID were investigated. Then 10 selected VMAT dose plans were measured by EPID with the rotational method. The overall variation of EPID dosimetric characteristics was within 1.4% for VMAT. The film system serving as a conventional tool for verification showed good agreement both with EPID measurements ([94.1 ± 1.5]% with 3 mm/3% criteria) and treatment planning system (TPS) calculations ([97.4 ± 2.8]% with 3 mm/3% criteria). In addition, EPID measurements for VMAT presented good agreement with TPS calculations ([99.1 ± 0.6]% with 3 mm/3% criteria). The EPID system performed the robustness of potential error findings in TPS calculations and the delivery system. This study demonstrated that an EPID system can be used as a reliable and efficient quality assurance tool for VMAT dose verification.

61 RADIATION PROTECTION AND DOSIMETRY↗

Standoff Spectroscopy of Surface Adsorbed Chemicals

Despite its immediate applications, selective detection of trace quantities of surface adsorbed chemicals, such as explosives, without physically collecting the sample molecules is a challenging task. Standoff spectroscopic techniques offer an ideal method of detecting chemicals without using a sample collection step. Though standoff spectroscopic techniques are capable of providing high selectivity, their demonstrated sensitivities are poor. Here we describe standoff detection of trace quantities of surface adsorbed chemicals using two quantum cascade lasers operated simultaneously, with tunable wavelength windows that match with absorption peaks of the analytes. This standoff method is a variation of photoacoustic spectroscopy, where scattered light from the sample surface is used for exciting acoustic resonance of the detector. We demonstrate a sensitivity of 100 ng/cm{sup 2} and a standoff detection distance of 20 m for surface adsorbed analytes such as explosives and tributyl phosphate.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗

Role of f Electrons in the Optical and Photoelectrochemical Behavior of Ca(La 1– x Ce x ) 2 S 4 (0 ≤ x ≤ 1)

This study focuses on a solid solution series, Ca(La 1– x Ce x ) 2 S 4 (0 ≤ x ≤ 1), where the f electron density is absent in CaLa 2 S 4 and is progressively increased until it is maximized in CaCe 2 S 4 . Correspondingly, these samples, synthesized by a sealed ampule method, showed progressive variations in color ranging from gray for CaLa 2 S 4 to orange-red for CaCe 2 S 4 . The crystal structural nuances of both the end members and three solid solutions with $x$ = 0.25, 0.50, and 0.75 were established with the complementary use of synchrotron X-ray diffraction and neutron scattering. Interestingly, these data were consistent with a two-phase composition centered around each nominal solid solution stoichiometry. Optical characterization via diffuse reflectance spectroscopy and Tauc analyses showed a shrinking of the energy band gap (from the UV to vis range) when Ce was progressively introduced into the host CaLa 2 S 4 structure. Overall, these data were in concert with electronic band structure calculations, using density functional theory, which showed the progressive formation of an intermediate f band when Ce was introduced intro the structure. Photoelectrochemical measurements in an aqueous redox electrolyte, as well as surface photovoltage and Kelvin probe measurements, revealed all samples to be n-type semiconductors. The valence and conduction band edge positions of the end members and the three solid solutions could be mapped, on both the redox and vacuum reference energy scales, by combining these measurements with the optical data.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗

SchrödingerNet: A Universal Neural Network Solver for the Schrödinger Equation

Recent advances in machine learning have facilitated numerically accurate solution of the electronic Schrödinger equation (SE) by integrating various neural network (NN)-based wave function ansatzes with variational Monte Carlo methods. Nevertheless, such NN-based methods are all based on the Born–Oppenheimer approximation (BOA) and require computationally expensive training for each nuclear configuration. In this work, we propose a novel NN architecture, SchrödingerNet, to solve the full electronic-nuclear SE by defining a loss function designed to equalize local energies across the system. This approach is based on a translationally, rotationally and permutationally symmetry-adapted total wave function ansatz that includes both nuclear and electronic coordinates. Furthermore, this strategy not only allows for an efficient and accurate generation of a continuous potential energy surface at any geometry within the well-sampled nuclear configuration space, but also incorporates non-BOA corrections, through a single training process. Comparison with benchmarks of atomic and small molecular systems demonstrates its accuracy and efficiency.

Chemical calculations↗

One-Step Relativistic Driven Similarity Renormalization Group Multireference Perturbation Theory

We present an efficient implementation of a one-step relativistic second-order multireference perturbation theory based on the multireference driven similarity renormalization group (MR-DSRG) using the exact two-component (X2C) Hamiltonian, which we denote X2C-DSRG-MRPT2. We show that the X2C-DSRG-MRPT2 method can accurately capture spin–orbit coupling (SOC) effects in the electronic structure of strongly correlated systems containing elements across the periodic table. We further demonstrate that the X2C-DSRG-MRPT2 method, through its variational treatment of SOC effects, can yield spin–orbit splittings with mean absolute percentage errors consistently below 7% with respect to experimental values for systems containing up to sixth row elements. With its modest computational scaling (fourth power in system size for the perturbative step) and high accuracy, X2C-DSRG-MRPT2 provides a promising avenue for the routine treatment of relativistic effects in strongly correlated molecular systems.

Hamiltonians↗

Unveiling Morphology and Crystallinity Dynamics in Ni x Mn 1– x CO 3 Cathode Precursors through Batch-Mode Coprecipitation

This study delves into the synthesis and control of Ni x Mn 1–x CO 3 , a critical class of Mn-rich, Co-free precursors vital for cathode-oxide materials in energy storage and conversion technologies. Employing batch-mode coprecipitation, we systematically generated samples with varying Ni concentrations (x = 0, 0.1, 0.3, 0.5, 0.7, and 0.9) and conducted a comprehensive analysis of their compositions, crystallinities, transition-metal distributions, and particle morphologies through both experimental and computational methods. A significant variation in particle size and crystallinity was observed, contingent on the Ni content. Further, a pivotal transition emerged at Ni concentrations above x = ~0.5, transforming uniform morphologies, such as spherical, monodisperse, pseudo-single-crystalline particles, into bimodal, polycrystalline structures. Furthermore, the study highlights the role of Ni–ammonia complexes leading to Ni-deficient precipitates and underscores the importance of ammonia concentration in achieving precise Ni content control. This study unveils critical reaction conditions governing Mn-rich precursor properties that are vital for cathode-oxides, emphasizing the need for meticulous synthetic control and offering the potential for practical applications in advanced energy storage and conversion systems.

25 ENERGY STORAGE↗

Data Imbalance, Uncertainty Quantification, and Transfer Learning in Data‐Driven Parameterizations: Lessons From the Emulation of Gravity Wave Momentum Transport in WACCM

Abstract Neural networks (NNs) are increasingly used for data‐driven subgrid‐scale parameterizations in weather and climate models. While NNs are powerful tools for learning complex non‐linear relationships from data, there are several challenges in using them for parameterizations. Three of these challenges are (a) data imbalance related to learning rare, often large‐amplitude, samples; (b) uncertainty quantification (UQ) of the predictions to provide an accuracy indicator; and (c) generalization to other climates, for example, those with different radiative forcings. Here, we examine the performance of methods for addressing these challenges using NN‐based emulators of the Whole Atmosphere Community Climate Model (WACCM) physics‐based gravity wave (GW) parameterizations as a test case. WACCM has complex, state‐of‐the‐art parameterizations for orography‐, convection‐, and front‐driven GWs. Convection‐ and orography‐driven GWs have significant data imbalance due to the absence of convection or orography in most grid points. We address data imbalance using resampling and/or weighted loss functions, enabling the successful emulation of parameterizations for all three sources. We demonstrate that three UQ methods (Bayesian NNs, variational auto‐encoders, and dropouts) provide ensemble spreads that correspond to accuracy during testing, offering criteria for identifying when an NN gives inaccurate predictions. Finally, we show that the accuracy of these NNs decreases for a warmer climate (4 × CO 2 ). However, their performance is significantly improved by applying transfer learning, for example, re‐training only one layer using ∼1% new data from the warmer climate. The findings of this study offer insights for developing reliable and generalizable data‐driven parameterizations for various processes, including (but not limited to) GWs.

54 ENVIRONMENTAL SCIENCES↗