SEARCH · Engineering Papers
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.
Causal inference methods are important to complement predictive machine learning models to improve theoretical understanding of the underlying system in climate science. Granger causality (GC) can determine whether time series value of one variable (e.g., X) can predict the future values of another variable (e.g., Y), but not the other way around.
Advection-dominated dynamical systems, characterized by partial differential equations, are found in applications ranging from weather forecasting to engineering design where accuracy and robustness are crucial. There has been significant interest in the use of techniques borrowed from machine learning to reduce the computational expense and/or improve the accuracy of predictions for these systems. These rely on the identification of a basis that reduces the dimensionality of the problem and the subsequent use of time series and sequential learning methods to forecast the evolution of the reduced state. Often, however, machine-learned predictions after reduced-basis projection are plagued by issues of stability stemming from incomplete capture of multiscale processes as well as due to error growth for long forecast durations. To address these issues, we have developed a non-autoregressive time series approach for predicting linear reduced-basis time histories of forward models. In particular, we demonstrate that non-autoregressive counterparts of sequential learning methods such as long short-term memory (LSTM) considerably improve the stability of machine-learned reduced-order models. Further, we evaluate our approach on the inviscid shallow water equations and show that a non-autoregressive variant of the standard LSTM approach that is bidirectional in the principal component directions obtains the best accuracy for recreating the nonlinear dynamics of partial observations. Moreover-and critical for many applications of these surrogates-inference times are reduced by three orders of magnitude using our approach, compared with both the equation-based Galerkin projection method and the standard LSTM approach.
Metal d-electron oxides having an odd number of electrons per cell should exhibit band degeneracy at the Fermi energy, making them, in band theory, formally metallic. In many cases, however, these are “false metals”, as evidenced by the observation that many ABO 3 oxide perovskites with a magnetic 3d B-atom are observed to be insulators both below and above the Néel temperature. These inconsistencies between experimental observation and expectation have historically been resolved by invoking degeneracy-breaking physics, based largely on -pure electron effects, such as strong interelectronic correlation for d-electron compounds (the Mott mechanism). Such explanations generally consider microscopic lattice or magnetic degrees of freedom (m-DOF) as largely passive spectators, not a cause of the formally metallic being an insulator. Yet, it has long been known that ABO 3 perovskites can manifest an arrangement of m-DOFs in the form of octahedral tilting, bond dimerization, Jahn-Teller distortions, and ordering of local magnetic moments. It appears reasonable that such structural and magnetic local degrees of freedom need to be allowed to compete with purely electronic strong correlation. To answer this question, we explored a range of d-electron oxide perovskites exemplified by the archetypes LaTiO 3 , LaVO 3 , SrMnO 3 , and LaMnO 3 with 1, 2, 3, or 4 d -electrons, respectively. Using a mean-field-like electronic structure method (here, density functional theory), we find that a combination of magnetic symmetry breaking with structural distortions can account for insulating band gaps in this series while at the same time correctly predicting for the control case an intrinsic paramagnetic metal in SrVO 3 as symmetry-breaking is insufficiently strong to remove the degeneracy. Finally, this indicates that calculating quantitatively local magnetic and positional symmetry-breaking motifs in unit cells that avoid averaging at the outset over the low symmetry motifs can provide consistent trends in a Mott transition without Mott U.
Analysis of methanol pool fire conducted as part of validation study for SIERRA/Fuego. Results showed trends & errors consistent with related studies. Area validation metric provides way to quantify model form uncertainty. AVM shows that more work could be done to understand how model form uncertainty varies with mesh resolution. There is a possible atypical use of MAVM on time-series data. AVM shows mismatch between predicted flame height and experimental value less sensitive to variations in mixture fraction than temperature. Mismatch about experimental value also more symmetric for mixture fraction. Our analysis showed that mixture fraction is preferable for this application.
Numerical weather forecasting models and statistical methods have found wide use to help power companies estimate renewable output, but better methods are needed, particularly for extended forecasts. Machine learning approaches have been used here as well, but so far a major limitation is the ability to also predict the corresponding uncertainty in a forecast. Here we show that both can be done and demonstrate this using a long-term-short memory neural network where the difference between predicted and ground truth data are used to train a model for the corresponding forecast uncertainties.
This is the final technical report for the SETO-funded VRN3P project (PNNL# 76914). The goal of this project, led by Pacific Northwest National Laboratory (PNNL), in collaboration with Lawrence Livermore National Laboratory (LLNL) and Portland General Electric (PGE), was to develop and validate a deep variational recurrent neural network-based net-load prediction (VRN3P) framework for probabilistic time-series forecasting of day-ahead net-load under high solar penetration scenarios. The project team reports successful design of a novel probabilistic net-load forecasting architecture, comprising of a variational autoencoder and a recurrent neural network, which demonstrates 30% improvement in forecast performance, 60% improvement in training time, and consumes 44% less memory, when compared with conventional baseline models. The team tested the VRN3P model performance on GridLAB-D test-cases representing varying BTM solar penetration levels of 20%, 30%, and 50%, with integrated time-series net-load profiles provided by the utility partner (PGE). The VRN3P model demonstrate <2% hourly MAPE (averaged over the year) for day- ahead net-load forecast on the test scenario with 20% BTM solar. Transfer learning extension of the VRN3P model has demonstrated 8.33× speed-up in training, while still achieving acceptable forecast performance of 2.24% hourly MAPE on the 30% BTM solar penetration test-scenario. A preliminary version of the VRN3P GridAPPS-D™has been developed, along with a web-based interactive user-interface (named ‘Forte’) which has made available on GitHub for public use.
An application of digital time series analysis to flutter test data processing was conducted. A numerical investigation was used to evaluate the method, as well as its sensitivity to noise and parameter variations. These parameters included those involved with data acquisition, as well as system response characteristics. This digital time series method was then used to predict flutter speed from subcritical response wind tunnel tests. Flutter speeds predicted from forced response, subcritical wind tunnel tests were compared to the experimental flutter speeds.
The recent discoveries of extrasolar planetary systems around lower main-sequence stars such as tau Booe (HD 120136) has prompted further investigation into their stellar activity. A cursory analysis of tau Booe for cyclic chromospheric activity, based on its 30-yr record of Ca 2 H and K fluxes obtained as part of the HK Project from Mount Wilson Observatory, finds an intermediate, sub-yearly period (approximately 117 d) in chromospheric activity in addition to, and separate from, both its rotation (3.3 d) and long-term variability. As a persistent subyearly period in surface magnetic activity is unprecedented, we investigate this apparent anomaly further by examining chromospheric activity levels of other stars with similar mass, searching for variability in chromospheric activity with periods of less than one year, but longer than measured or predicted rotation. An examination of the time series of 40 mid-to-late F dwarfs yielded one other star for further analysis: alpha Com (HD 114378, P approximately 132 d). The variations for these two stars were checked for persistence and coherence. Based on these determinations, we eliminate the possibilities of rotation, long-term activity cycle, and the evolution of active regions as the cause of this variation in both stars. In particular, for tau Booe we infer that the phenomenon may be chromospheric in origin; however, beyond this, it is difficult to identify anything further regarding the cause of the activity variations, or even whether the observed modulation in the two stars have the same origin.
Fire emissions of gases and aerosols alter atmospheric composition and have substantial impacts on climate, ecosystem function, and human health. Warming climate and human expansion in fire‐prone landscapes exacerbate fire impacts and call for more effective management tools. Here we developed a global fire forecasting system that predicts monthly emissions using past fire data and climate variables for lead times of 1 to 6 months. Using monthly fire emissions from the Global Fire Emissions Database (GFED) as the prediction target, we fit a statistical time series model, the Autoregressive Integrated Moving Average model with eXogenous variables (ARIMAX), in over 1,300 different fire regions. Optimized parameters were then used to forecast future emissions. The forecast system took into account information about region‐specific seasonality, long‐term trends, recent fire observations, and climate drivers representing both large‐scale climate variability and local fire weather. We cross‐validated the forecast skill of the system with different combinations of predictors and forecast lead times. The reference model, which combined endogenous and exogenous predictors with a 1 month forecast lead time, explained 52% of the variability in the global fire emissions anomaly, considerably exceeding the performance of a reference model that assumed persistent emissions during the forecast period. The system also successfully resolved detailed spatial patterns of fire emissions anomalies in regions with significant fire activity. This study bridges the gap between the efforts of near‐real‐time fire forecasts and seasonal fire outlooks and represents a step toward establishing an operational global fire, smoke, and carbon cycle forecasting system.
NASA’s Prediction of Worldwide Energy Resource (POWER) project facilitates the use of NASA Earth Science data holdings within the energy, agricultural, and building heating/cooling design industries. POWER packages solar and meteorological data at various temporal levels from several NASA projects in a user friendly GIS-enabled web services system (https://power.larc.nasa.gov). Data users can access these data either through an intuitive data viewer, image services fully integrable with GIS analysis, connections in the cloud through an Amazon Web Services S3 Bucket, or fully customizable access through an API. Data provided by POWER has been used to remotely monitor solar array fields and integrated in a sizing tool for off-grid solar and storage systems. POWER data has also been coupled with key building decision tools to support design and retrofitting of building energy systems for energy efficiency and reduction of greenhouse gases. POWER is now developing capabilities to provide time series of the projected future evolution of surface quantities important to future energy production and use, such as heating/cooling degree days, temperature, wind speed, and downwelling solar flux. We present here a range of possible future changes in these quantities at locations throughout the continental United States. We show how both average and extreme values of the quantities will evolve from present-day to future climate conditions. We plan to provide these projections for users in the energy and sustainable energy communities.
Linear prediction theory for multivariate, one- dimensional, stationary, stochastic processes with finite pasts
Time series of methane fluxes measured by eddy-covariance require gap-filling to estimate annual emissions. Gap-filling methane fluxes is challenging because of high variability and complex responses to multiple drivers. To date, there is no widely established gap-filling standard for methane, with regards both to the best model algorithms and predictors. In this study, we address the need for standardization by synthesizing results of gap-filling methods applied at 17 wetland sites spanning boreal to tropical regions including all major wetlands classes and two rice paddies. We introduce new procedures for: 1) creating realistic artificial gap scenarios, 2) training and evaluating gap-filling models without overstating performance, and 3) predicting half-hourly methane fluxes and annual emissions with robust uncertainty estimates. We tested a conventional method (marginal distribution sampling) and four machine learning algorithms - penalized linear regression, artificial neural networks, random forests, and boosted decision trees - and four predictor sets, including temporal, meteorological, ecosystem carbon and energy flux, and soil predictors. We find that the conventional method can achieve similar median performance to the machine learning models but is worse than the best machine learning models and relatively insensitive to predictor choices. Of the machine learning models, decision tree algorithms performed the best in cross-validation experiments, even with a baseline predictor set, and artificial neural networks showed comparable performance when using all predictors. Soil temperature was frequently the most important predictor whilst water table depth was important at sites with substantial water table fluctuations, highlighting the value of data on soil conditions. Raw gap-filling uncertainties from the machine learning models were underestimated and we propose a method to calibrate uncertainties to observations. Finally, we gap-fill and provide summary evaluation metrics for all 81 sites in the FLUXNET-CH4 community dataset and publicly release the python code for model development, evaluation, and uncertainty estimation.
Shrub fractional abundance (SFA), the proportion of shrub cover per unit area, serves as a critical indicator of environmental aridity and ecosystem health in arid and semi-arid regions, particularly across the Mongolian steppe. However, large-scale SFA mapping in Mongolian steppe ecosystems remains challenging due to the small crown size of shrubs, their sparse distribution, and spectral overlap with coexisting low vegetation (e.g., grasses and herbs), which hinders accurate detection using coarser-resolution satellite data or traditional field surveys. To address these challenges, we developed a two-step approach that integrates very-high-resolution (VHR) imagery, time-series Sentinel-2 data, and deep learning techniques. First, we generated high-accuracy benchmark maps of individual shrub crowns from 0.5 m VHR imagery by combining manual segmentation with a hybrid deep learning framework (Dino V2 and convolutional neural networks). Second, we used these shrub crown maps as training data to build an XGBoost model for predicting SFA from 20 m Sentinel-2 time-series data, leveraging phenological information to improve estimation. We validated our approach across 70 sites (1km 2 each) in the Inner Mongolia Autonomous Region, which is representative of Mongolian steppe ecosystems. From VHR imagery, we mapped 1.31 million shrub crowns with an accuracy of R 2 = 0.92. Scaling up with Sentinel-2 data yielded regional SFA maps with an R 2 = 0.60. Further SHAP (SHapley Additive exPlanations) analysis on the developed XGBoost model revealed that phenological metrics (particularly observations in early-May, mid-July, and late-September), which distinguish shrub phenology from that of other land cover types (e.g., grasses and bare soil), were the most influential predictors of SFA. Finally, our regional SFA maps uncovered unimodal relationships between shrub distribution and climate variables, peaking at mean annual minimum temperatures near 0 °C and annual precipitation around 200 mm. Collectively, these findings demonstrate how the integration of multi-source remote sensing and machine learning can overcome historical limitations in SFA mapping, enabling accurate, spatially continuous assessments across vast Inner-Mongolian steppe ecosystems. Our framework has the potential to be applied to other steppe ecosystems and dryland ecosystems across the Mongolian steppe and beyond, offering a foundation for improved monitoring and ecological impact assessments in the face of global climate changes.
This paper addresses the problem of building trust in the online prediction of a battery powered aircraft's remaining flying time. A series of flight tests is described that make use of a small electric powered unmanned aerial vehicle (eUAV) to verify the performance of the remaining flying time prediction algorithm. The estimate of remaining flying time is used to activate an alarm when the predicted remaining time is two minutes. This notifies the pilot to transition to the landing phase of the flight. A second alarm is activated when the battery charge falls below a specified limit threshold. This threshold is the point at which the battery energy reserve would no longer safely support two repeated aborted landing attempts. During the test series, the motor system is operated with the same predefined timed airspeed profile for each test. To test the robustness of the prediction, half of the tests were performed with, and half were performed without, a simulated powertrain fault. The pilot remotely engages a resistor bank at a specified time during the test flight to simulate a partial powertrain fault. The flying time prediction system is agnostic of the pilot's activation of the fault and must adapt to the vehicle's state. The time at which the limit threshold on battery charge is reached is then used to measure the accuracy of the remaining flying time predictions. Accuracy requirements for the alarms are considered and the results discussed.
We present a selection of attempts to infer geomagnetic storm properties from time series data. Initially, we unsuccessfully attempt to predict the duration of a geomagnetic storm based on data preceding the onset of a geomagnetic storm. Similar techniques are also used to attempt to predict the maximum depth of a geomagnetic storm. Additionally, we present a case study in Gaussian Process regression, where we are able to accurately interpolate between geomagnetic storm parameters, but are unable to predict future storm behavior. Lastly, we describe attempts at using random forest regression with sliding windows to infer geomagnetic storm behavior at future times. These results did not provide robust forecasts of geomagnetic storm behavior, but did occasionally correctly predict a storm within the margin of error. Techniques presented in this paper include machine learning classification and regression, Gaussian Process regression, and machine learning regression with sliding windows. Many of the attempts presented in this document did not result in meaningful predictions of geomagnetic storm behavior. We hope this document can be used as a guide for future readers attempting to predict geomagnetic storm properties.
KOGUT — Knowledge Oriented Graph Unified Transformer KOGUT implements the Relational Graph Transformer (RelGT) architecture for knowledge graph link prediction in biological domains, with a primary focus on microbial growth media prediction. While the original RelGT (arXiv:2505.10960) targets relational tables, time series, and multi-table databases, KOGUT adapts this architecture for heterogeneous biological knowledge graphs, providing first-in-class AI predictive models for microbial cultivation. Key Adaptations Beyond Original RelGT: - Knowledge Graph Focus: Applied to biological KGs with semantic node types (taxa, chemicals, media, phenotypes, environments) versus generic relational database tables, trained on the KG-Microbe knowledge graph (1.3M entities, 2.9M edges, 24 relation types). - Multimodal Node Encoding: Integrates node labels, categories, descriptions, and synonyms from KG metadata through learned embedding layers—adapting relational column features to graph node attributes with textual semantics. - Extended K-Hop Subgraph Strategy: Optimized neighborhood sampling (3-hop default, configurable up to 200 nodes) tuned for sparse biological networks, building on the original local-global attention framework with biological relation preservation. - Biolink Predicate Preservation: Type-specific transformations for 24 biological edge semantics (occurs_in, consumes, produces, has_phenotype, subclass_of) beyond standard relational foreign keys, enabling multi-relation link prediction. - Inductive Learning Support: Enables zero-shot predictions for novel taxa through feature-based embeddings (temperature, oxygen requirements, gram stain, cell shape), extending the original transductive relational benchmark scope to uncultured microorganisms. CheapSOTA Performance Optimizations (This Distribution): - VQ-EMA Centroid Attention: Vector quantization with exponential moving average for improved global context modeling (+5-10% MRR improvement). - HDF5 Precomputed Data Loading: One-time preprocessing of k-hop subgraphs to eliminate redundant graph traversals (2-5× training speedup). - Distributed Data Parallel Training: Multi-GPU support for scaling to larger knowledge graphs (tested on 4× NVIDIA A100 GPUs at NERSC Perlmutter). - Mixed Precision Training: Automatic mixed precision (AMP) for memory efficiency and faster training. Advantages Over Standard Knowledge Graph Embedding Models: Combines RelGT's proven multi-element tokenization (features, type, hop, structure) with graph-native biological representations, enabling interpretable link prediction across heterogeneous entities that standard embedding models (TransE, RotatE, ComplEx) and table-based transformers cannot directly model. Achieves near-perfect performance on microbial growth media prediction (MRR: 0.9966, Precision@1: 0.9932, Hit@10: 1.0000) while maintaining explainability through attention-based reasoning over biological pathways. Training Data: - KG-Microbe merged knowledge graph: 1,379,337 nodes, 2,960,472 edges - 24 biological relation types including taxonomic hierarchies, metabolic interactions, phenotype associations, and environmental relationships - Primary prediction task: Growth media suitability for microbial taxa (biolink:occurs_in, 50K edges) - Multi-relation capability: Predicts links for any of the 24 relation types, including chemical consumption/production, phenotype associations, and taxonomic classification Citation: Original RelGT Architecture: Dwivedi et al., "Relational Graph Transformer", arXiv:2505.10960, 2025 KOGUT Implementation: Knowledge Oriented Graph Unified Transformer for Microbial Growth Media Prediction Developed at Lawrence Berkeley National Laboratory (LBNL) Trained on NERSC Perlmutter supercomputer
Empirical tests of a particular least squares multiple prediction program were studied, as well as the relationship of this particular program to general least squares multiple prediction theory. Empirical evaluation and test of the program involved: (1) conversion of the program to run on an IBM 360/44 with replication of test results obtained on a Univac 1108; and (2) generation and analysis of Monte Carlo simulated data with the objective of comparison against results theoretically obtainable from spectrum analysis routines.
Episodes of forest mortality have been observed worldwide associated with climate change, impacting species composition and ecosystem services such as water resources and carbon sequestration. Yet our ability to predict forest mortality remains limited, especially across large scales. Time series of satellite imagery has been used to document ecosystem resilience globally, but it is not clear how well remotely sensed resilience can inform the prediction of forest mortality across continental, multi-biome scales. Here, we leverage forest inventories across the continental United States to systematically assess the potential of ecosystem resilience derived using different data sets and methods to predict forest mortality. Here we found high resilience was associated with low mortality in eastern forests but was associated with high mortality in western regions. The unexpected resilience–mortality relation in western United States may be due to several factors including plant trait acclimation, insect population dynamics, or resource competition. Overall, our results not only supported the opportunity to use remotely sensed ecosystem resilience to predict forest mortality but also highlighted that ecological factors may have crucial influences because they can reverse the sign of the resilience–mortality relationships.