Search NASA⌕ Search

SEARCH · Search NASA

Results for “MODEL SIMULATION DATA”

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 865 records · Page 48

Investigation of realistic PET simulations incorporating tumor patient's specificity using anthropomorphic models: Creation of an oncology database

Purpose: The GATE Monte Carlo simulation toolkit is used for the implementation of realistic PET simulations incorporating tumor heterogeneous activity distributions. The reconstructed patient images include noise from the acquisition process, imaging system's performance restrictions and have limited spatial resolution. For those reasons, the measured intensity cannot be simply introduced in GATE simulations, to reproduce clinical data. Investigation of the heterogeneity distribution within tumors applying partial volume correction (PVC) algorithms was assessed. The purpose of the present study was to create a simulated oncology database based on clinical data with realistic intratumor uptake heterogeneity properties.Methods: PET/CT data of seven oncology patients were used in order to create a realistic tumor database investigating the heterogeneity activity distribution of the simulated tumors. The anthropomorphic models (NURBS based cardiac torso and Zubal phantoms) were adapted to the CT data of each patient, and the activity distribution was extracted from the respective PET data. The patient-specific models were simulated with the Monte Carlo Geant4 application for tomography emission (GATE) in three different levels for each case: (a) using homogeneous activity within the tumor, (b) using heterogeneous activity distribution in every voxel within the tumor as it was extracted from the PET image, and (c) using heterogeneous activity distribution corresponding to the clinical image following PVC. The three different types of simulated data in each case were reconstructed with two iterations and filtered with a 3D Gaussian postfilter, in order to simulate the intratumor heterogeneous uptake. Heterogeneity in all generated images was quantified using textural feature derived parameters in 3D according to the ground truth of the simulation, and compared to clinical measurements. Finally, profiles were plotted in central slices of the tumors, across lines with heterogeneous activity distribution for visual assessment.Results: The accuracy of the simulated database was assessed against the original clinical images. The PVC simulated images matched the clinical ones best. Local, regional, and global features extracted from the PVC simulated images were closest to the clinical measurements, with the exception of the size zone variability and the mean intensity values, where heterogeneous tumors showed better reproducibility. The profiles on PVC simulated tumors after postfiltering seemed to represent the more realistic heterogeneous regions with respect to the clinical reference.Conclusions: In this study, the authors investigated the input activity map heterogeneity in the GATE simulations of tumors with heterogeneous activity distribution. The most realistic heterogeneous tumors were obtained by inserting PVC activity distributions from the clinical image into the activity map of the simulation. Partial volume effect (PVE) can play a crucial role in the quantification of heterogeneity within tumors and have an important impact on applications such as patient follow-up during treatment and assessment of tumor response to therapy. The development of such a database incorporating patient anatomical and functional variability can be used to evaluate new image processing or analysis algorithms, while providing control of the ground truth, which is not available when dealing with clinical datasets. The database includes all images used and generated in this study, as well as the sinograms and the attenuation phantoms for further investigation. It is freely available to the interested reader of the journal at http://www.med.upatras.gr/oncobase/.

62 RADIOLOGY AND NUCLEAR MEDICINE↗

Nonparametric reconstruction of the dark energy equation of state

The major aim of ongoing and upcoming cosmological surveys is to unravel the nature of dark energy. In the absence of a compelling theory to test, a natural approach is to first attempt to characterize the nature of dark energy in detail, the hope being that this will lead to clues about the underlying fundamental theory. A major target in this characterization is the determination of the dynamical properties of the dark energy equation of state w. The discovery of a time variation in w(z) could then lead to insights about the dynamical origin of dark energy. This approach requires a robust and bias-free method for reconstructing w(z) from data, which does not rely on restrictive expansion schemes or assumed functional forms for w(z). We present a new non parametric reconstruction method for the dark energy equation of state based on Gaussian Process models. This method reliably captures nontrivial behavior of w(z) and provides controlled error bounds. We demollstrate the power of the method on different sets of simulated supernova data. The GP model approach is very easily extended to include diverse cosmological probes.

71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSIC↗

Relating Convective and Stratiform Rain to Latent Heating

The relationship among surface rainfall, its intensity, and its associated stratiform amount is established by examining observed precipitation data from the Tropical Rainfall Measuring Mission (TRMM) Precipitation Radar (PR). The results show that for moderate-high stratiform fractions, rain probabilities are strongly skewed toward light rain intensities. For convective-type rain, the peak probability of occurrence shifts to higher intensities but is still significantly skewed toward weaker rain rates. The main differences between the distributions for oceanic and continental rain are for heavily convective rain. The peak occurrence, as well as the tail of the distribution containing the extreme events, is shifted to higher intensities for continental rain. For rainy areas sampled at 0.58 horizontal resolution, the occurrence of conditional rain rates over 100 mm/day is significantly higher over land. Distributions of rain intensity versus stratiform fraction for simulated precipitation data obtained from cloud-resolving model (CRM) simulations are quite similar to those from the satellite, providing a basis for mapping simulated cloud quantities to the satellite observations. An improved convective-stratiform heating (CSH) algorithm is developed based on two sources of information: gridded rainfall quantities (i.e., the conditional intensity and the stratiform fraction) observed from the TRMM PR and synthetic cloud process data (i.e., latent heating, eddy heat flux convergence, and radiative heating/cooling) obtained from CRM simulations of convective cloud systems. The new CSH algorithm-derived heating has a noticeably different heating structure over both ocean and land regions compared to the previous CSH algorithm. Major differences between the new and old algorithms include a significant increase in the amount of low- and midlevel heating, a downward emphasis in the level of maximum cloud heating by about 1 km, and a larger variance between land and ocean in the new CSH algorithm.

Tao, Wei-Kuo↗

Modeling of time-lapse seismic monitoring using CO 2 leakage simulations for a model CO 2 storage site with realistic geology: Application in assessment of early leak-detection capabilities

Time-lapse surface seismic surveys have been widely used at carbon sequestration sites for site characterization, monitoring subsurface CO 2 plume migration, and detecting potential CO 2 leakage from a storage reservoir. Monitoring in the first permeable unit directly above the primary seal is important for early detection of CO 2 leakage. Forward modeling of time-lapse seismic data can be used to assess the utility of the seismic method for CO 2 leakage detection. We develop a workflow for forward modeling of time-lapse seismic data, including constructing seismic velocity models using flow simulation outputs, modeling of pre-stack and post-stack synthetic seismic data following seismic data processing sequence and analysis of processed synthetic time-lapse seismic data. We apply the forward modeling and analysis workflow to assessing the detectability and the earliest detection time of seismic monitoring using the hypothetical CO 2 leakage scenarios for a model geologic storage site with realistic geology. We then derive the detection thresholds using the simulated normalized root-mean-square (NRMS) differences for the no-leakage case at a range of signal-to-noise ratios, representing the background noise levels in seismic data. We then compare NRMS differences triggered by the CO 2 leakage to the detection thresholds at each time step to quantify the detectability and the earliest detection time of seismic monitoring. We analyze the effects of the acquisition parameters and elastic parameters on the produced synthetic seismic data and earliest detection time. Our modeling results indicate that high signal-to-noise ratio is needed to detect the CO 2 leakage at the model site. Minimizing the background noise in seismic data is crucial for improving the detectability of the seismic method. Increasing the shot density or increasing the dominant frequency of the source wavelet is likely to increase the possibility of the leakage detection and reduce the time needed for the detection. The elastic parameters used in the rock physics modeling have significant effects on the resultant seismic velocity models and synthetic seismic data, highlighting their critical roles in understanding and interpreting time-lapse seismic reflection data associated with CO 2 monitoring, verification and accounting activities.

58 GEOSCIENCES↗

Predicting carbon dioxide and energy fluxes across global FLUXNET sites with regression algorithms

Spatio-temporal fields of land–atmosphere fluxes derived from data-driven models can complement simulations by process-based land surface models. While a number of strategies for empirical models with eddy-covariance flux data have been applied, a systematic intercomparison of these methods has been missing so far. In this study, we performed a cross-validation experiment for predicting carbon dioxide, latent heat, sensible heat and net radiation fluxes across different ecosystem types with 11 machine learning (ML) methods from four different classes (kernel methods, neural networks, tree methods, and regression splines). We applied two complementary setups: (1) 8-day average fluxes based on remotely sensed data and (2) daily mean fluxes based on meteorological data and a mean seasonal cycle of remotely sensed variables. The patterns of predictions from different ML and experimental setups were highly consistent. There were systematic differences in performance among the fluxes, with the following ascending order: net ecosystem exchange ( R 2 < 0.5), ecosystem respiration ( R 2 > 0.6), gross primary production ( R 2 > 0.7), latent heat ( R 2 > 0.7), sensible heat ( R 2 > 0.7), and net radiation ( R 2 > 0.8). The ML methods predicted the across-site variability and the mean seasonal cycle of the observed fluxes very well ( R 2 > 0.7), while the 8-day deviations from the mean seasonal cycle were not well predicted ( R 2 < 0.5). Fluxes were better predicted at forested and temperate climate sites than at sites in extreme climates or less represented by training data (e.g., the tropics). Finally, the evaluated large ensemble of ML-based models will be the basis of new global flux products.

54 ENVIRONMENTAL SCIENCES↗

Characterization of imaging performance in differential phase contrast CT compared with the conventional CT: Spectrum of noise equivalent quanta NEQ(k)

Purpose: Differential phase contrast CT (DPC-CT) is emerging as a new technology to improve the contrast sensitivity of conventional attenuation-based CT. The noise equivalent quanta as a function over spatial frequency, i.e., the spectrum of noise equivalent quanta NEQ(k), is a decisive indicator of the signal and noise transfer properties of an imaging system. In this work, we derive the functional form of NEQ(k) in DPC-CT. Via system modeling, analysis, and computer simulation, we evaluate and verify the derived NEQ(k) and compare it with that of the conventional attenuation-based CT. Methods: The DPC-CT is implemented with x-ray tube and gratings. The x-ray propagation and data acquisition are modeled and simulated through Fresnel and Fourier analysis. A monochromatic x-ray source (30 keV) is assumed to exclude any system imperfection and interference caused by scatter and beam hardening, while a 360 Degree-Sign full scan is carried out in data acquisition to avoid any weighting scheme that may disrupt noise randomness. Adequate upsampling is implemented to simulate the x-ray beam's propagation through the gratings G{sub 1} and G{sub 2} with periods 8 and 4 {mu}m, respectively, while the intergrating distance is 193.6 mm (1/16 of the Talbot distance). The dimensions of the detector cell for data acquisition are 32 Multiplication-Sign 32, 64 Multiplication-Sign 64, 96 Multiplication-Sign 96, and 128 Multiplication-Sign 128 {mu}m{sup 2}, respectively, corresponding to a 40.96 Multiplication-Sign 40.96 mm{sup 2} field of view in data acquisition. An air phantom is employed to obtain the noise power spectrum NPS(k), spectrum of noise equivalent quanta NEQ(k), and detective quantum efficiency DQE(k). A cylindrical water phantom at 5.1 mm diameter and complex refraction coefficient n= 1 -{delta}+i{beta}= 1 -2.5604 Multiplication-Sign 10{sup -7}+i1.2353 Multiplication-Sign 10{sup -10} is placed in air to measure the edge transfer function, line spread function and then modulation transfer function MTF(k), of both DPC-CT and the conventional attenuation-based CT. The x-ray flux is set at 5 Multiplication-Sign 10{sup 6} photon/cm{sup 2} per projection and observes the Poisson distribution, which is consistent with that of a micro-CT for preclinical applications. Approximately 360 regions, each at 128 Multiplication-Sign 128 matrix, are used to calculate the NPS(k) via 2D Fourier transform, in which adequate zero padding is carried out to avoid aliasing in noise. Results: The preliminary data show that the DPC-CT possesses a signal transfer property [MTF(k)] comparable to that of the conventional attenuation-based CT. Meanwhile, though there exists a radical difference in their noise power spectrum NPS(k) (trait 1/|k| in DPC-CT but |k| in the conventional attenuation-based CT) the NEQ(k) and DQE(k) of DPC-CT and the conventional attenuation-based CT are in principle identical. Conclusions: Under the framework of ideal observer study, the joint signal and noise transfer property NEQ(k) and detective quantum efficiency DQE(k) of DPC-CT are essentially the same as those of the conventional attenuation-based CT. The findings reported in this paper may provide insightful guidelines on the research, development, and performance optimization of DPC-CT for extensive preclinical and clinical applications in the future.

62 RADIOLOGY AND NUCLEAR MEDICINE↗

SNNPG: Using Spiking Neural Networks to Detect Attacks in the Power Grid

We explore the potential of Spiking Neural Networks (SNN) to enhance the security of power grid operations by detecting False Data Injection (FDI) attacks. These attacks manipulate PMU readings, leading to erroneous control decisions and grid disruptions. We develop a method to convert Phase Measurement Unit (PMU) data into spike trains, capturing both temporal and spatial dimensions. Using an SNN model, we conduct evaluations with simulated power grid data, showcasing accuracy in detecting FDI attacks. SNN models rapidly identify anomalies in real-time PMU data, safeguarding grid operations by alerting operators to irregular readings and preventing incorrect decisions.

artificial intelligence↗

Gas Gun Model and Comparison to Experimental Performance of Pipe Guns Operating with Light Propellant Gases and Large Cryogenic Pellets

Injection of multiple large (~10 to 30 mm diameter) shattered pellets into ITER plasmas is presently part of the scheme planned to mitigate the deleterious effects of disruptions on the vessel components. To help in the design and optimize performance of the pellet injectors for this application, a model referred to as “the gas gun simulator” has been developed and benchmarked against experimental data. The computer code simulator is a Java program that models the gas-dynamics characteristics of a single-stage gas gun. Following a stepwise approach, the code utilizes a variety of input parameters to incrementally simulate and analyze the dynamics of the gun as the projectile is launched down the barrel. Using input data, the model can calculate gun performance based on physical characteristics, such as propellant-gas and fast-valve properties, barrel geometry, and pellet mass. Although the model is fundamentally generic, the present version is configured to accommodate cryogenic pellets composed of H2, D2, Ne, Ar, and mixtures of them and light propellant gases (H2, D2, and He). The pellets are solidified in situ in pipe guns that consist of stainless steel tubes and fast-acting valves that provide the propellant gas for pellet acceleration (to speeds ~200 to 700 m/s). The pellet speed is the key parameter in determining the response time of a shattered pellet system to a plasma disruption event. The calculated speeds from the code simulations of experiments were typically in excellent agreement with the measured values. With the gas gun simulator validated for many test shots and over a wide range of physical and operating parameters, it is a valuable tool for optimization of the injector design, including the fast valve design (orifice size and volume) for any operating pressure (~40 bar expected for the ITER application) and barrel length for any pellet size (mass, diameter, and length). Key design parameters and proposed values for the pellet injectors for the ITER disruption mitigation systems are discussed.

Combs, Stephen Kirk↗

A Review and Comparison of Mouse and Rat Responses to Micro Gravity, Hyper Gravity and Simulated Models of Partial Gravity; Species Differences, Gaps in the Available Data, and Consideration of the Advantages and Caveats of Each Model for Spaceflight

Laboratory strains of mice and rat are widely used to study mammalian responses to stimulus, and both have been studied under a variety of gravity conditions, including space flight. We compared results obtained from exposure to spaceflight and microgravity, hyper gravity via centrifugation, earth gravity, and models of simulated partial gravity (hind-limb unloading and partial weight bearing treatments). We examined the reported changes in survival, body mass, circadian rhythm (body temperature and activity levels), behavior, bone, muscle, immune, cardio-vasculature, vestibular, reproduction and neonate survival, microbiome, and the visual system. Not all categories have published data for both species, some have limited data, and there are variations in experiment design that allow for only relative comparisons to be considered. The data reveal species differences in both the level of gravity required to obtain a response, degree of response, and in temporal expression of responses. Examination of the data across the gravity levels allows consideration of the hypothesis that gravitational responses follow a continuum, and organ specific differences are noted. In summary, we present advantages and caveats of each model system as pertains to gravitational biology research and identify gaps in our knowledge of how these mammals respond to gravity.

mouse rat gravity↗

Bayesian Calibration of Inertial Confinement Fusion Models with Very Expensive Data

Calibrating computer simulations against experimental data is a ubiquitous problem in science and is particularly difficult when experimental data are expensive or diffi- cult to collect. In inertial confinement fusion experiments, where a single data point can cost 10 5 –10 6 USD, calibration of large-scale design calculations is particularly challenging. We present a framework that aims to provide calibrated predictions far from the available data that can be trained using a small number of data points. The approach aims to retain as much of the physics (summarized by the simulation code) as possible and is based on a truncated polynomial chaos expansion in the space of simulation inputs, augmented with a non-parametric correction for higher distribution modes. Due to the difficulty of the problem, understanding the sources of uncertainty in the model is essential; we demonstrate this through an application of our approach to a series of 6 experimental laser shots. As a result, we validate our approach, show that the uncertainties in our predictions of experimental performance are dominated by shot-shot fluctuations and regression uncertainties, and discuss paths towards an improved prediction model.

97 MATHEMATICS AND COMPUTING↗

DiaMonD: An Integrated Multifaceted Approach to Mathematics at the Interfaces of Data, Models, and Decisions

DiaMonD is a multi-institutional DOE MMICC effort that aims to address the challenges of end-to-end, data-to-decisions modeling and simulation for complex computational science and engineering problems in a unified, systematic, and integrated fashion. Institutions involved are Colorado State, Florida State, Los Alamos, MIT, Oak Ridge, Stanford, and UT-Austin. The goals of DiaMonD are (1) to develop advanced mathematical methods and analysis for multimodel, multiphysics, multiscale model problems driven by frontier DOE applications, including those in subsurface energy and environmental flows, materials for energy storage and conversion, and ice sheet dynamics; (2) to create theory and algorithms for integrated inversion, optimization, and uncertainty quantification for these complex problems; and (3) to disseminate the philosophy of an integrated end-to-end, data-to-decisions approach to modeling and simulation of complex problems to the broader applied math and computational science communities through workshops and other outreach.

42 ENGINEERING↗

Prospects for distinguishing dark matter models using annual modulation

It has recently been demonstrated that, in the event of a putative signal in dark matter direct detection experiments, properly identifying the underlying dark matter-nuclei interaction promises to be a challenging task. Given the most optimistic expectations for the number counts of recoil events in the forthcoming Generation 2 experiments, differentiating between interactions that produce distinct features in the recoil energy spectra will only be possible if a strong signal is observed simultaneously on a variety of complementary targets. However, there is a wide range of viable theories that give rise to virtually identical energy spectra, and may only differ by the dependence of the recoil rate on the dark matter velocity. In this work, we investigate how degeneracy between such competing models may be broken by analyzing the time dependence of nuclear recoils, i.e. the annual modulation of the rate. For this purpose, we simulate dark matter events for a variety of interactions and experiments, and perform a Bayesian model-selection analysis on all simulated data sets, evaluating the chance of correctly identifying the input model for a given experimental setup. Lastly, we find that including information on the annual modulation of the rate may significantly enhance the ability of a single target to distinguish dark matter models with nearly degenerate recoil spectra, but only with exposures beyond the expectations of Generation 2 experiments.

72 PHYSICS OF ELEMENTARY PARTICLES AND FIELDS↗

The Circulation of the Plasmasphere Fluid during the Erosion Event on September 8, 2017

A strong solar wind pressure pulse triggered the magnetic storm on September 7, 2017. Near the end of September 7, the z-component of the interplanetary magnetic field (IMF Bz) dropped from 9 to -10 nT in 30 min. The IMF Bz remained at the level of -10 nT for 2 hours and then had another rapid drop to -31 nT in 30 min. The sudden plunge of IMF Bz and the associated strong convection electric field stirred up the storm main phase with Dst falling from 0 to -122 nT from 2200 UT on September 7 to 0200 UT on the 8th. Severe plasmasphere erosion was observed on September 8 by multiple spacecraft, such as the Van Allen Probes and the Arase satellite. In this study, we examine the fate of the eroded plasmasphere particles during this event by model simulation as well as satellite data analysis. The simulation tool we use is the Space Weather Modeling Framework (SWMF)/Block-Adaptive Tree Solarwind Roe-type Upwind Scheme (BATS-R-US) model coupled with the Comprehensive Inner Magnetosphere-Ionosphere (CIMI) model. One of the distinctive capabilities of the SWMF/BATSRUS-CIMI model is that it treats the cold plasmas in the plasmasphere as a separate fluid in the MHD equations. As a result, the transport and circulation of the plasmasphere fluid in the global magnetosphere can be traced and the impacts of this cold fluid on the global magnetosphere can be evaluated. In this paper, we will show how the drainage plume is formed during the storm and how the plasmasphere fluid is transported to the flank and lobe regions and eventually to the plasma sheet and reenters into the plasmasphere. We will validate our simulation by plasmasphere signatures observed in both the inner and outer magnetosphere.

Mei-Ching Fok↗

Full event interpretation with machine-learning-based particle-flow reconstruction in the CMS detector

The particle-flow (PF) algorithm constructs a global description of each particle collision by producing a comprehensive list of final-state particles, and is central to event reconstruction in the CMS experiment at the CERN LHC. The existing PF implementation relies on physics-motivated heuristics and assumptions that can be replaced by machine-learning (ML) models trained directly on simulated data and naturally suited to modern graphics processing units (GPUs). A state-of-the-art ML-based PF (MLPF) reconstruction algorithm, implemented within the CMS software framework, is presented. The MLPF algorithm performs a learnable full-event reconstruction on GPUs, generalizes across detector conditions and collision energies, and replaces multiple modular reconstruction steps with a single unified model. Physics performance comparable to standard PF reconstruction is achieved in both simulation and data, with improved jet energy resolution and inference time. In simulated top quark-antiquark events under LHC Run-3 (2023-2024) conditions, the jet energy resolution improves by 10-20% for jets with transverse momentum between 30-100 GeV. Inference time is evaluated using simulated multijet events, with a median of $20\,\hbox {ms}$ per event on an Nvidia L4 GPU, compared to approximately $110\,\hbox {ms}$ for the standard CMS PF reconstruction.

Hayrapetyan, Aram [Yerevan Phys. Inst.]↗

Model-based parameter estimation using cardiovascular response to orthostatic stress

This paper presents a cardiovascular model that is capable of simulating the short-term (< or approximately equal to 3 min) transient hemodynamic response to gravitational stress and a gradient-based optimization method that allows for the automated estimation of model parameters from simulated or experimental data. We perform a sensitivity analysis of the transient heart rate response to determine which parameters of the model impact the heart rate dynamics significantly. We subsequently include only those parameters in the estimation routine that impact the transient heart rate dynamics substantially. We apply the estimation algorithm to both simulated and real data and showed that restriction to the 20 most important parameters does not impair our ability to match the data.

Non-NASA Center↗

A model-data intercomparison of CO2 exchange across North America: Results from the North American Carbon Program site synthesis

Our current understanding of terrestrial carbon processes is represented in various models used to integrate and scale measurements of CO2 exchange from remote sensing and other spatiotemporal data. Yet assessments are rarely conducted to determine how well models simulate carbon processes across vegetation types and environmental conditions. Using standardized data from the North American Carbon Program we compare observed and simulated monthly CO2 exchange from 44 eddy covariance flux towers in North America and 22 terrestrial biosphere models. The analysis period spans 220 site-years, 10 biomes, and includes two large-scale drought events, providing a natural experiment to evaluate model skill as a function of drought and seasonality. We evaluate models' ability to simulate the seasonal cycle of CO2 exchange using multiple model skill metrics and analyze links between model characteristics, site history, and model skill. Overall model performance was poor; the difference between observations and simulations was 10 times observational uncertainty, with forested ecosystems better predicted than nonforested. Model-data agreement was highest in summer and in temperate evergreen forests. In contrast, model performance declined in spring and fall, especially in ecosystems with large deciduous components, and in dry periods during the growing season. Models used across multiple biomes and sites, the mean model ensemble, and a model using assimilated parameter values showed high consistency with observations. Models with the highest skill across all biomes all used prescribed canopy phenology, calculated NEE as the difference between GPP and ecosystem respiration, and did not use a daily time step.

0793↗

Machine Learned Empirical Numerical Integrator from Simulated Data

Recently, a number of state-of-the-art surrogate machine learning (ML) models have been designed for global weather and climate prediction, which have been trained using reanalysis data products. Reanalysis data products are constructed using numerical model simulations that combine numerical integration of partial differential equations and parameterization schemes. These products are typically only archived and made available using coarsened spatial and temporal resolutions. This study explores the impact of the numerical generation methods used to produce the training datasets and the temporal resolution of those datasets on machine learning surrogate models. Using the nonlinear vector autoregression (NVAR) machine as an explainable ML technique, simple dynamical systems are emulated with ML models trained on data produced by three classical numerical integration schemes. NVAR is validated as a skillful ML method, capable of producing accurate predictions and, more importantly, reconstructing both the underlying dynamics and the numerical integration scheme used to generate the training data. However, the machine fails to generalize predictions on unseen test data generated by different numerical integration schemes, despite the underlying dynamical system being the same. This result provides a word of caution for the growing field of machine learning emulation of weather and climate dynamics. Furthermore, we illustrate using NVAR that training on temporally coarsened data may increase the required complexity of ML models and potentially introduce new numerical challenges. Finally, we discover that empirical integration schemes with arbitrary time-stepping sizes can be constructed directly from the data, which implies a potential for the development of empirical numerical integration schemes.

54 ENVIRONMENTAL SCIENCES↗