Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “validation dataset”

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 919 records · Page 51
All sourcesNASA NTRSDOE OSTI
DOE OSTI2024

Empirical Validation of UBEM: An Assessment of Bias in Urban Building Energy Modeling for Chicago

Residential and commercial buildings currently account for 30% of total global final energy consumption. Urban-scale building energy modeling (UBEM) can enable scalable investments and unlock building improvements by quantifying energy, demand, emissions, and cost reductions of specific measures or packages for building-specific technologies in large geographic regions. While the sophistication of UBEM data sources and technologies have increased dramatically in the past decade, there remains a knowledge gap for empirical validation and sources of bias between building-specific energy models and measured data at varying geographic scales.As UBEM continues to develop, systemic analysis of accuracy, bias, and limitations of the resulting models is necessary to inform best practices and move toward standardization. These are characterized for the Automatic Building Energy Modeling (AutoBEM) software suite with an initial case study involving metered electricity consumption data from 247,188 buildings in Chicago, Illinois, USA - averaged across years 2019-2021 - compared to the following datasets: (1) the AutoBEM-generated nation-scale Model America version 2 (MAv2) data for 596,064 buildings, (2) tax assessor data for 579,829 buildings, (3) tax assessor data filled with MAv2, and (4) 102 representative dynamic archetypes. The accuracy is reported for every building type and vintage combination, along with multiple sources of bias for unique building descriptors. The AutoBEM simulation workflow produced energy consumption estimates that closely match aggregated metered electricity consumption data for different types of buildings constructed during various time periods at the city scale - with initial normalized mean bias error of 10.9%, and 1.1% after removing outliers. Contribution of statistically significant factors including building type, land use, age, and size to variance in UBEM bias is quantified.

Garg, Ankur↗
DOE OSTI2026

A hybrid Penman-Monteith and machine learning model for simulating evapotranspiration and its components

Integrating physical processes with machine learning has advanced evapotranspiration (ET) simulation, yet most hybrid models fail to partition total ET into its components: soil evaporation (E) and vegetation transpiration (T). This study introduces Residual Neural Network–Penman–Monteith (RNN-PM), a novel hybrid dual-source ET model designed to overcome this limitation. The model synergizes the physically-based Penman–Monteith framework with three specialized residual neural networks trained to estimate key conductance parameters (canopy conductance, soil surface conductance, and aerodynamic conductance). Furthermore this explicit parameterization allows for the direct partitioning of total ET. Validation at National Ecological Observatory Network (NEON) flux sites using high-frequency partitioned E and T shows that RNN-PM reliably reproduces ET and the transpiration fraction (T/ET). For ET, the model achieves an average Kling–Gupta efficiency (KGE) of 0.89 and a root-mean-square error (RMSE) of 0.55 mm/day; for T/ET, the KGE is 0.87 with an RMSE of 0.06. Furthermore, RNN-PM demonstrates robust generalization, accurately simulating ET and its components well beyond the initial training dataset, even under extreme climatic conditions. This study extended the analysis by comparing the RNN-PM model with seven established dual-source ET models. The results indicate that RNN-PM outperforms both conventional machine learning models and purely physical process-based models in simulating ET components in most cases. Among the purely physical process-based dual-source models, those based on surface temperature decomposition showed improved performance as the leaf area index (LAI) decreased when evaluated against high-frequency ET component datasets. In contrast, the performance of conductance-based dual-source models declined with decreasing LAI. Although purely machine learning-based models can produce relatively accurate simulations of ET components, they often exhibit limited generalization capability, an issue that the RNN-PM model effectively overcomes. Ultimately, the RNN-PM model represents a significant advance in simulating ET components, offering a novel and scalable approach for improving the representation of land–atmosphere interactions in Earth system models.

54 ENVIRONMENTAL SCIENCES↗
DOE OSTI2025

A Variational Autoencoder Model Toward Molecular Structure Representation Learning of Fuels

Here, in this work, a Variational Autoencoder (VAE)-based data-driven modeling framework is developed with the overarching goal of enabling fuel design. The VAE model is trained on a large dataset with several chemical species to learn a compressed latent space molecular representation. Chemical structure in the form of Simplified Molecular Input Line Entry System (SMILES) string is fed as input, encoded into the VAE latent space, and decoded back to the SMILES string using Long Short-Term Memory (LSTM) networks. Complexities of the VAE training loss function are thoroughly examined by varying the weightage (beta (𝜷) parameter) of the latent space regularization term, thereby assessing the balance between reconstruction accuracy and validity, and focusing on both accurate molecular structure reconstruction and latent space consistency. Two different strategies for 𝜷 variation are evaluated: linear annealing and cyclic annealing. In addition, the impact of total correlation adjustment and hierarchical priors is also studied with regard to the balance between reconstruction fidelity and latent space regularization, and potential issues such as posterior collapse, over-regularization, and poor disentanglement of latent variables. Overall, the best performance of the model is achieved with hierarchical priors and incrementally increasing 𝜷 from 0 to a threshold value of 0.25 over 75 epochs. The generative VAE model can be readily coupled with Quantitative Structure–Property Relationship (QSPR) analysis to develop an integrated end-to-end framework for fuel-property prediction and molecular design of novel promising fuels.

fuel design↗
DOE OSTI2026

Random forest models accurately classify synthetic opioids using high-dimensionality mass spectrometry datasets

Detection of novel threat agents presents several challenges, a principle one being the development of untargeted methods to screen an increasing number of threat chemicals whose exact structures are unknown. With the use of Machine Learning (ML) tools, we can guide the development of analytical methods for broad-spectrum detection of unbounded threat chemical families in complex mixtures. Toward this goal, we used nominal mass and high-resolution mass spectrometry data for hundreds of synthetic opioids and non-opioid compounds. We tested two ML techniques, logistic regression and random forest, to develop models towards a practical, implementable method for opioid detection. We found that of these tested ML methods, random forest models resulted in the highest validation accuracy (95+%) for both nominal mass and high-resolution classification of opioids versus non-opioids, with low false positive and false negative rates. The RF models were then used to successfully predict the classification of 10 compounds—five opioids and five non-opioids not part of the training and validation analysis. This application of ML is a critical step towards the development of field-deployable nominal mass spectrometers with ML-driven analyses for classification of emergent threats.

Chemistry↗
DOE OSTI2025

Fast physics-based launcher optimization for electron cyclotron current drive

With the increased urgency to design fusion pilot plants, fast optimization of electron cyclotron current drive (ECCD) launchers is paramount. Traditionally, this is done by coarsely sampling the 4D parameter space of possible launch conditions consisting of (1) the launch location (constrained to lie along the reactor vessel), (2) the launch frequency, (3) the toroidal launch angle, and (4) the poloidal launch angle. For each initial condition, a ray-tracing simulation is performed to evaluate the ECCD efficiency. Unfortunately, this approach often requires a large number of simulations (sometimes millions in extreme cases) to build up a dataset that adequately covers the plasma volume, which must then be repeated every time the design point changes. Here we adopt a different approach. Rather than launching rays from the plasma periphery and hoping for the best, we instead directly reconstruct the optimal ray for driving current at a given flux surface using a reduced physics model coupled with a commercial ray-tracing code. Repeating this throughout the plasma volume requires only hundreds of simulations, constituting a significant speedup. The new method is validated on two separate example tokamak profiles, and is shown to reliably drive localized current at the specified flux surface with the same optimal efficiency as obtained from the traditional approach.

70 PLASMA PHYSICS AND FUSION TECHNOLOGY↗
DOE OSTI2025

Random forest models accurately classify synthetic opioids using high-dimensionality mass spectrometry datasets

Detection of novel threat agents presents several challenges, a principle one being the development of untargeted methods to screen an increasing number of threat chemicals whose exact structures are unknown. With the use of Machine Learning (ML) tools, we can guide the development of analytical methods for broad-spectrum detection of unbounded threat chemical families in complex mixtures. Toward this goal, we used nominal mass and high-resolution mass spectrometry data for hundreds of synthetic opioids and non-opioid compounds. We tested two ML techniques, logistic regression and random forest, to develop models towards a practical, implementable method for opioid detection. We found that of these tested ML methods, random forest models resulted in the highest validation accuracy (95+%) for both nominal mass and high-resolution classification of opioids versus non-opioids, with low false positive and false negative rates. The RF models were then used to successfully predict the classification of 10 compounds—five opioids and five non-opioids not part of the training and validation analysis. This application of ML is a critical step towards the development of field-deployable nominal mass spectrometers with ML-driven analyses for classification of emergent threats.

Arasteh, Kourosh [Lawrence Livermore National Labo↗
DOE OSTI2025

Preliminary Design of Engineering-Scale Salt Accident Analysis Facility to Support Molten Salt Reactor Licensing

Systems-level nuclear accident analysis codes for reactor licensing must be validated using experimental data that represent behaviors expected during actual full-scale accidents. Some behaviors may arise from coupled processes and only manifest at large scales. This report presents the preliminary design of the Salt Accident Analysis Facility (SAAF, pronounced “safe”), which is an experimental test facility to be constructed at Argonne that can be used to conduct integrated salt accident tests at an engineering scale. The measurement capabilities of the SAAF are based on previously developed methods and will provide the representative datasets that are needed to support molten salt reactor (MSR) licensing. Details of the design, the processes to be quantified, the measurement techniques for quantifying the processes, the variables that can be adjusted to simulate different accident scenarios, and operational considerations are presented herein. This report provides stakeholders the opportunity to give feedback on the test facility capabilities and planned analyses before it is constructed.

22 GENERAL STUDIES OF NUCLEAR REACTORS↗
DOE OSTI2023

Global normal spectral irradiance in Albuquerque - Data and Resources

The Photovoltaic Systems Evaluation Laboratory (PSEL) at Sandia National Laboratories (SNL) in Albuquerque, NM has measured global normal spectral irradiance nearly continuously from August 2013 to April 2018. During this time other broadband irradiance measurements (global horizontal, direct normal, diffuse horizontal and global normal) and weather variables were also recorded. For this dataset PV Performance Labs (PVPL) has pulled together data from both sources to assemble a full calendar year spectral data set for use in photovoltaic research. It is composed of eight continuous segments of different durations taken from the two-year period September 2013 to August 2015.

photovoltaics↗
DOE OSTI2020

Quantifying the generalization error in deep learning in terms of data distribution and neural network smoothness

We report the accuracy of deep learning, i.e., deep neural networks, can be characterized by dividing the total error into three main types: approximation error, optimization error, and generalization error. Whereas there are some satisfactory answers to the problems of approximation and optimization, much less is known about the theory of generalization. Most existing theoretical works for generalization fail to explain the performance of neural networks in practice. To derive a meaningful bound, we study the generalization error of neural networks for classification problems in terms of data distribution and neural network smoothness. We introduce the cover complexity (CC) to measure the difficulty of learning a data set and the inverse of the modulus of continuity to quantify neural network smoothness. A quantitative bound for expected accuracy/error is derived by considering both the CC and neural network smoothness. Although most of the analysis is general and not specific to neural networks, we validate our theoretical assumptions and results numerically for neural networks by several data sets of images. The numerical results confirm that the expected error of trained networks scaled with the square root of the number of classes has a linear relationship with respect to the CC. We also observe a clear consistency between test loss and neural network smoothness during the training process. In addition, we demonstrate empirically that the neural network smoothness decreases when the network size increases whereas the smoothness is insensitive to training dataset size.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2020

Comprehensive Detection of Single Amino Acid Variants and Evaluation of Their Deleterious Potential in a PANC-1 Cell Line

Identifying single amino acid variants (SAAVs) in cancer is critical for precision oncology. Several advanced algorithms are now available to identify SAAVs but attempts to combine different algorithms and optimize them on large datasets to achieve a more comprehensive coverage of SAAVs have not been implemented. Herein we report an expanded detection of SAAVs in the PANC-1 cell line using three different strategies, which results in identification of 540 SAAVs in the mass spectrometry data, the most comprehensive coverage yet reported. Among the set of 540 SAAVs, 79 are evaluated as deleterious SAAVs based on analysis using novel AssVar software in which one of the driver mutations found in each protein of KRAS, TP53 and SLC37A4 is further validated using independent selected reaction monitoring (SRM) analysis. Our study represents the most comprehensive discovery of SAAVs to date and the first large-scale detection of deleterious SAAVs in the PANC-1 cell line. This work may serve as the basis for future research in pancreatic cancer and personal immunotherapy and treatment.

59 BASIC BIOLOGICAL SCIENCES↗
DOE OSTI2025

UBW (USLCI-Brightway2) [SWR-25-169]

Life cycle inventory (LCI) data are critical for robust life cycle assessment (LCA), yet many widely used datasets such as the U.S. Life Cycle Inventory (USLCI) are not natively compatible with advanced modeling frameworks like Brightway2. This work presents an automated pipeline to transform USLCI data into a fully functional Brightway2 project. The workflow performs systematic data cleaning, resolves duplicate process and exchange identifiers, and applies allocation to multi-output processes. Technosphere and biosphere flows are harmonized through unit conversions and a bridge mapping to the biosphere3 database, with comprehensive logging of missing flows and cutoff issues. The resulting Brightway2 database is validated using matrix diagnostics to ensure consistency of the technosphere, and is benchmarked via life cycle impact assessment (LCIA) methods such as ReCiPe and IPCC GWP. Outputs include reproducible CSV exports of corrected processes, elementary flows, characterization factors, and LCIA results, alongside backup utilities for project sharing. This pipeline lowers barriers for integrating USLCI data into open-source LCA workflows, enabling reproducible, validated LCA inventories within the Brightway 2 framework.

Ghosh, Tapajyoti [National Laboratory of the Rocki↗
DOE OSTI2023

Evaluating the Interplay between Trajectory Segmentation and Mode Inference Error

Travel behavior changes are essential to transportation decarbonization. Travel diaries, consisting of sequences of trips between places, are typically used to instrument human travel behavior. However, these diaries are only as accurate as the underlying methods used to construct them. Travel diary algorithms have been a popular research topic since the advent of Global Positioning System tracking surveys. These algorithms have typically been validated using prompted recall of presegmented trips, thus disregarding the continuity of mode inference. Phone operating systems have adopted battery-conserving techniques, but the resulting data collection errors have not been studied extensively. We introduce a framework to evaluate the accuracy of trip length computations and mode inference by analyzing continuous mode-segmented trajectories for groups of trips. We then use the framework to identify the input data quality and the impact of postprocessing. Our primary inputs to this evaluation are MobilityNet, a public dataset containing information from three artificial timelines covering 15 different travel modes, and sample open-source travel diary creation algorithms from the OpenPATH project. Our framework concretely shows that the variance of the distance error drops from (0.217, 0.0848) to (o.011, 0.0407) (Android, iOS) after postprocessing. Similarly, the weighted F-scores for mode inference increase from (0.25, 0.29) to (0.60, 0.74) (iOS, Android) between random forest and geographic information system-based models. We hope that this standardized method will be adapted to evaluate other, potentially proprietary, travel diary algorithms. Finally, the results can be used to understand and improve the state of the art in the travel diary creation field.

33 ADVANCED PROPULSION SYSTEMS↗
DOE OSTI2023

Integrated Design of Ultradurable, Low CO 2 Alternative Binder Systems via Machine Learning

This ARPA-E project developed a machine learning tool to use in formulation design of cementitious binders for concrete having 50% less embodied CO 2 and possessing twice the durability compared to concrete based on ordinary portland cement (OPC) binders. The technical focus was on limestone/calcined clay cement (LC3), the leading replacement for OPC. Here, hierarchical machine learning (HML) was used to model the flowability, set time, strength, and durability of LC3 concrete. This methodology identifies latent variables derived from domain knowledge and empirical models that develop an accurate model for a response surface from small datasets. For the flowability metric, particle packing was a dominant factor, while strength and durability were both strongly determined by the fraction of metakaolin and the water:solids ratio. Under constraints of water:binder ratio, material performance metrics, embodied CO 2 , and cost per tonne of OPC, multi-objective optimization was used to design binders parameterized by the mineral composition replacing OPC, particle size distributions, and water:solids ratio. The trained algorithm was able to predict multiple mixes met these performance criteria, and experimental testing validated the predictions. The model demonstrated here is relevant for North America, where pure kaolin deposits are found broadly. The approach is being taken forward into commercial application by Ansatz AI, a materials informatics company founded by PI Washburn and co-PI Poczos. Through collaborations with the cement and concrete industry, and funding from SBIR programs, a commercial software will be developed in future research.

36 MATERIALS SCIENCE↗
DOE OSTI2022

Progress Towards a Predictive Eagle Behavior and Risk Modeling Framework: Overview and Recent Validation Efforts

This presentation summarizes progress to date of the U.S. Department of Energy project, "Development of a computational framework for modeling golden eagles (Aquila chrysaetos) near wind farms," which focuses on stochastic behavioral modeling of soaring raptors across landscape, facility, and turbine spatiotemporal scales. This publicly available, open-source modeling framework includes behavioral models based on three different underlying principles: energy minimization at landscape scale, behavioral heuristics at landscape-facility scale, and data-driven behaviors at the facility-micro-scale. We will briefly overview the key advancements in the behavioral modeling state of the art, which leverages multiple high-resolution telemetry data sources combined with high-fidelity atmospheric flow modeling insights. We then present preliminary results from a validation study in Altamont, California. This new study involves a novel application of the Stochastic Soaring Raptor Simulator (SSRS), in a new geographic locale, to understand facility scale eagle movement patterns over time scales representative of a wind project's lifetime. For this desktop analysis (that does not depend on any high-performance computing resources), SSRS simultaneously considers a variety of wind conditions and eagle approach vectors toward a project site of interest. This work demonstrates the integration of publicly available landscape-scale atmospheric datasets, our recently improved engineering updraft models (see presentation from Thedin et al.), and our energy minimization behavioral models within the SSRS framework. While we only present results from a single behavioral model, the integration of these three modeling components forms the foundation for our more sophisticated behavioral models (see presentations from Brandes et al., Sandhu et al.) that are under active development. Results are presented in the form of presence maps, which may be applied to estimate risk to wildlife, augment ground survey data, inform wind-plant operations, or incorporated into wind-plant designs.

agent-based modeling↗
DOE OSTI2022

Hawaii Wave Surge Energy Converter (HAWSEC) OSU O.H. Hinsdale Basin

The following information and metadata applies to both the Phase I (Hydrodynamics) and Phase II (Full System Power Take-Off) zip folders which contain testing data from the OSU (Oregon State University) O.H. Hinsdale Wave Research Laboratory, from both OSU and the University of Hawaii at Manoa (UH). See zip folders provided further below in the downloads section. For experimental data of the full system, including PTO, see Phase II dataset. There are two main directories in each Phases's zip folder: "OSU_data" and "UH_data". The "OSU_data" directory contains data collected from their DAQ (data acquisition system), which includes all wave gauge observations, as well as body motions derived from their Qualisys motion tracking system. The organization of the directory follows OSU's convention. Detailed information on the instrument setup can be found under "OSU_data/docs/setup/instm_locations". The experiments conducted are documented in the "OSU_data/docs/daq_logs", which provides the trial number to the corresponding data located under "OSU_data/data" in several formats (e.g., ".mat" and ".txt"). Inside the trial directory, data is provided for each of the instruments defined in "OSU_data/docs/setup/instm_locations". The "UH_data" directory contains data collected from their DAQ. The data is stored in a ".tdms" file format. There are free plug-ins for Microsoft Excel and MathWorks MATLAB to read the ".tdms" format. Below are a few links providing methods to read in the data, but a Google search should identify alternatives sources if these no longer exist (valid as of January 2024): Excel: http://www.ni.com/example/27944/en/ MATLAB: https://www.mathworks.com/matlabcentral/fileexchange/30023-tdms-reader The Excel plugin is recommend to get a quick overview of the data. The UH data is organized by directory name, in which the sub-directories for each experiment contains a directory whose name defines the wave height and period for the experimental data within. For example, a directory name "H02_T0275" corresponds to an experiment with wave height 0.1m and a period of 2.75s. For random wave data, the gamma value is also included in the directory name. For example, a directory name "H02_T0225_G18" corresponds to an experiment with a significant wave height of 0.2m, a peak period of 2.25s, and a gamma value of 1.8, with each spectra being a TMA spectrum. For the free decay experiments, the directory name is defined by the initial angular displacement. For example, a directory name "ang05_run01" corresponds to an experiment with an initial angular displacement of 5 degrees. There is a dataset in the UH data for each corresponding experiment defined in the OSU DAQ logs. The ".tdms" data is output from the DAQ at fixed intervals. Therefore, if multiple files are contained within the folder, the data will need to be stitched together. Within the UH dataset, there are two input channels from the OSU DAQ providing a random square wave signal for time synchronization ("ENV-WHT-0010") and a high/low signal ("ENV-WHT-0012") to identify when the wave maker is active (+5V). The UH data is logged as a collection of channel outputs. Channels not in use for the OSU testing (either Phase I or Phase II) are marked "nan" below. If the sensor is disconnected, it will record noise throughout the experiment. Below are the channel definitions in terms of what they measure: GPS Time = time CYL-POS-0001 = position between flap and fixed reference CYL-LCA-0001 = force between flap and hydraulic cylinder REC-LPT-0001 = nan REC-HPT-0001 = nan REC-HPT-0002 = nan REC-HPT-0003 = nan HHT-HPT-0001 = pressure at exhaust ("head" only) REC-FQC-0001 = nan REC-FQC-0002 = nan HHT-FQC-0001 = flow at exhaust ("head" only) ENV-WHT-0001 = nan ENV-WHT-0002 = nan ENV-WHT-0003 = nan ENV-WHT-0010 = random signal from OSU DAQ ENV-WHT-0012 = high/low signal from OSU DAQ Also included is a calibration curve to convert the string pot data to flap pi...

16 TIDAL AND WAVE POWER↗
DOE OSTI2024

The use of digital thread for reconstruction of local fiber orientation in a compression molded pin bracket via deep learning

A deep convolutional neural network (DCNN) was used for microstructure reconstruction using artificial intelligence (MR-AI) by predicting local average fiber orientation distributions (FOD) in a 3D prepreg platelet molded composite (PPMC) pin bracket. To train the MR-AI model, surface strain fields from residual stresses simulated in PPMC plates were used as the input to the DCNN. A training dataset included PPMC plates with various degrees of global fiber alignment, based on the information obtained from high-fidelity flow simulation of a pin bracket. Further, the MR-AI model was then deployed to analyze FOD in the 3D pin bracket by conducting thermo-elastic residual stress analysis. Initially, the MR-AI model was established entirely on the synthetic simulation data. Then, a μCT scan of a physically molded pin bracket was used to create a finite element model that provided data for additional validation of the DCNN model. For the μCT scan finite element pin bracket the MR-AI model predicted the distribution of fiber orientation tensor components with MAE of 0.10 indicating a global prediction error of 10%. For the flow simulated pin bracket, the MR-AI model predicted the distribution of fiber orientation tensor components with a global prediction error of 11%. The MR-AI model showed the ability to predict regions of varying alignment in the base and flange of the pin bracket. The proposed MR-AI methodology allows for rapid prediction of FOD in geometrically complex parts and offers a promising path to detecting unique fiber orientation states in molded components.

42 ENGINEERING↗
DOE OSTI2025

Collisional Excitation of HCN by CO to Refine the Modeling of Cometary Comae

Here, we present the first dataset of collisional (de)-excitation rate coefficients of HCN induced by CO, one of the main perturbing gases in cometary atmospheres. The dataset spans the temperature range of 5–50 K. It includes both state-to-state rate coefficients involving the lowest ten and nine rotational levels of HCN and CO, respectively, and the so-called “thermalized” rate coefficients over the rotational population of CO at each kinetic temperature. The derivation of these coefficients exploited the good performance of the statistical adiabatic channel model (SACM) on top of an accurate interaction potential computed at the CCSD(T)-F12b/CBS level of theory. The reliability of the SACM approach was validated by comparison with full quantum calculations restricted at the lowest total angular momentum of the system. These results provide essential input to accurately model the distribution among the rotational energy levels and the abundance of HCN in cometary atmospheres, accounting for deviations from local thermodynamic equilibrium that typically occurs in such environments.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗
DOE OSTI2023

MetaFlux: Meta-learning global carbon fluxes from sparse spatiotemporal observations

We provide a global, long-term carbon flux dataset of gross primary production and ecosystem respiration generated using meta-learning, called MetaFlux. The idea behind meta-learning stems from the need to learn efficiently given sparse data by learning how to learn broad features across tasks to better infer other poorly sampled ones. Using meta-trained ensemble of deep models, we generate global carbon products on daily and monthly timescales at a 0.25-degree spatial resolution from 2001 to 2021, through a combination of reanalysis and remote-sensing products. Site-level validation finds that MetaFlux ensembles have lower validation error by 5–7% compared to their non-meta-trained counterparts. In addition, they are more robust to extreme observations, with 4–24% lower errors. We also checked for seasonality, interannual variability, and correlation to solar-induced fluorescence of the upscaled product and found that MetaFlux outperformed other machine-learning based carbon product, especially in the tropics and semi-arids by 10–40%. Overall, MetaFlux can be used to study a wide range of biogeochemical processes.

54 ENVIRONMENTAL SCIENCES↗