Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “Deep”

Search indexed NASA NTRS and DOE OSTI research on propulsion, heat transfer, battery materials and energy systems. Follow report and document links to the original sources.

Quote a phrase for an exact phrase match. Source license links do not imply unrestricted reuse.

At least 19 records
All sourcesNASA NTRSDOE OSTI
DOE OSTI2025

Deep Point Cloud Building Envelope Segmentation (DeeP-CuBES) using Deep Learning

Building Information Modeling (BIM) plays an important role in building design and construction, particularly for achieving energy-efficient retrofits. Building envelope retrofits using panelized prefabricated system, such as those popularized by the Energiesprong program, need accurate as-built dimensions of facade features (windows, doors, etc.) to achieve the desired thermal and air tightness. Traditionally, building surveying is done manually, resulting in a time-consuming and labor-intensive process. Recently, 3D point clouds from terrestrial LiDAR have been used to automate the generation of as-built dimensions of existing buildings. However, automated BIM using LiDAR relies on solving the point cloud semantic segmentation (PCSS) problem. In this work, we propose a robust pipeline for solving the PCSS problem using deep neural networks, focusing on overcoming challenges posed by imbalanced datasets and complex architectural features. We introduce the first high-density, labeled, and validated building envelope point cloud dataset derived from multiple building scans, specifically curated to tackle challenges in facade-level segmentation. Results from the trained neural networks show that advanced attention-based architectures and incorporating radiometry (light intensity and RGB) features significantly boost segmentation accuracy for windows and doors.

Selvakumar, Balaji [ORNL]↗
DOE OSTI2026

Are Deep Energy Retrofits in Commercial Buildings Including Window Upgrades?

U.S. Commercial buildings account for about 20% of total U.S. energy consumption. Because the thermal performance of windows significantly affects building energy efficiency and HVAC system performance, best practice guidance often includes window and envelope improvements in conjunction with HVAC upgrades to optimize energy use and improve occupant comfort. It is an open question, however, regarding how often these best practices are implemented in the field. This paper aims to address that gap by conducting a literature review and a series of interviews with commercial building auditing and management professionals to explore the factors that drive window retrofits in commercial buildings. The paper explores a range of case studies from deep energy retrofits across the globe, comparing projects with and without window retrofits. The primary goals of this review are to: (1) provide data from real-world case studies illustrating the role of windows in deep energy renovations and HVAC upgrades, (2) conduct retrofit cost analyses for windows and high-performance HVAC systems and (3) offer insights into how window upgrade decisions are made and when they are implemented as part of deep energy retrofits. Most of the retrofit studies focused exclusively on high performance HVAC upgrades without considering how window upgrades might further enhance the overall energy efficiency of commercial buildings. Interviews with building industry experts shed light on the key factors influencing deep energy retrofit decisions and what factors tip the scales in favor of including window measures with more comprehensive retrofit projects.

Cort, Katherine↗
DOE OSTI2026

Neural Scaling Laws of Deep ReLU and Deep Operator Network: A Theoretical Study

Neural scaling laws play a pivotal role in the performance of deep neural networks and have been observed in a wide range of tasks. However, a complete theoretical framework for understanding these scaling laws remains underdeveloped. In this paper, we explore the neural scaling laws for deep operator networks, which involve learning mappings between function spaces, with a focus on the Chen and Chen style architecture. These approaches, which include the popular Deep Operator Network (DeepONet), approximate the output functions using a linear combination of learnable basis functions and coefficients that depend on the input functions. We establish a theoretical framework to quantify the neural scaling laws by analyzing its approximation and generalization errors. We articulate the relationship between the approximation and generalization errors of deep operator networks and key factors such as network model size and training data size. Moreover, we address cases where input functions exhibit low-dimensional structures, allowing us to derive tighter error bounds. These results also hold for deep ReLU networks and other similar structures. Our results offer a partial explanation of the neural scaling laws in operator learning and provide a theoretical foundation for their applications.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2026

Data for FUN-PROSE: A Deep Learning Approach to Predict Condition-Specific Gene Expression in Fungi

mRNA levels of all genes in a genome is a critical piece of information defining the overall state of the cell in a given environmental condition. Being able to reconstruct such condition-specific expression in fungal genomes is particularly important to metabolically engineer these organisms to produce desired chemicals in industrially scalable conditions. Most previous deep learning approaches focused on predicting the average expression levels of a gene based on its promoter sequence, ignoring its variation across different conditions. Here we present FUN-PROSE—a deep learning model trained to predict differential expression of individual genes across various conditions using their promoter sequences and expression levels of all transcription factors. We train and test our model on three fungal species and get the correlation between predicted and observed condition-specific gene expression as high as 0.85. We then interpret our model to extract promoter sequence motifs responsible for variable expression of individual genes. We also carried out input feature importance analysis to connect individual transcription factors to their gene targets. A sizeable fraction of both sequence motifs and TF-gene interactions learned by our model agree with previously known biological information, while the rest corresponds to either novel biological facts or indirect correlations.

Genomics↗
DOE OSTI2025

Bayesian Optimized Deep Ensemble for Uncertainty Quantification of Deep Neural Networks: a System Safety Case Study on Sodium Fast Reactor Thermal Stratification Modeling

Deep neural networks (DNNs) are increasingly important to scientific computing and engineering system simulations. Accurate uncertainty quantification (UQ) for DNNs is critical in safety-sensitive engineering domains. Traditional Deep Ensemble (DE) methods, while easy to implement, frequently suffer from poorly calibrated uncertainty estimates and limited predictive accuracy due to reliance on fixed architectures with varied weight initializations. To address these issues, we introduce a workflow that combines Bayesian Optimization (BO) and DE. The workflow is modular, scalable, and integrates parallel BO initialized with Sobol sequences to individually optimize the hyperparameters of each ensemble member. This method enhances ensemble diversity, improves predictive accuracy, and provides reliable uncertainty estimates. We evaluate the proposed BODE approach in a sodium fast reactor thermal stratification modeling case study, where we used a densely connected convolutional neural network to predict turbulent viscosity during the reactor transient with consideration of data noise. We benchmark its performance against several optimization approaches, including baseline deep ensemble, evolutionary algorithm-optimized ensemble, ensemble formed via random search combined with greedy selection, and a BO ensemble using random initialization. Here, our results demonstrate superior performance of the developed BODE approach. In noise-free scenarios, BODE notably reduces incorrect aleatoric uncertainty and significantly enhances predictive accuracy. Under conditions of 5% and 10% Gaussian noise, BODE adaptively quantifies uncertainty proportional to data noise, achieving up to an 80% reduction in root mean square error compared to baseline methods and producing well-calibrated prediction intervals.

Bayesian optimization↗
DOE OSTI2026

How deep is your soil? Quantifying and spatially analyzing understudied deep soil in the United States

Deep soil is largely understudied and important in understanding biogeochemical processes in soil. Here, understudied soil is defined as the difference between soil studied to a known depth and the estimated bedrock depth. To understand more about deep soil, the understudied soil in the US was quantified and spatially analyzed using soil survey data and model estimates of bedrock depth. An equation was derived to find understudied soil using the dataset parameters “max lower depth studied”, “depth to bedrock”, and “likelihood of bedrock in the top 200 cm”. The survey data and bedrock model revealed that soil has been studied to an average depth of 1-2 meters, and the average depth to bedrock is 20 meters. Soil data density in the soil surveys was greatest in the West Coast, Midwest, and areas historically managed for agricultural, while the non-contiguous US and interior West were underrepresented. The soil had been studied deeper than the estimated soil depth in 455 out of 56,889 observation points concentrated in Alaska, California, Texas, Florida, Puerto Rico, and the US Virgin Islands. To understand the diversity and any taxonomic bias of the global soil data available, soil order was compared to US-based National Resource Conservation Service percentages and it was found that Oxisols, Alfisols, Ultisols, Andisols, and Histosols were overrepresented while Gelisols, Aridisols, Vertisols, Entisols, and Spodosols are underrepresented. Soil depth is important in exploring the complexity of biogeochemical processes that take place in soil.

Bedrock↗
DOE OSTI2026

Dependence of Deep Convective Cell Properties on Meteorological and Aerosol Conditions during TRACER

Deep convective cells significantly influence Earth’s energy balance and water cycle. However, their accurate representation in numerical models remains challenging due to their small spatiotemporal scales and limited observational constraints. This study examines over ∼400 deep convective cells near Houston, observed by a dual-polarization C-band radar during the Tracking Aerosol Convection Interactions Experiment (TRACER) intensive observation period (June–September 2022). Cells are categorized by lifetime into short-lived (<40 min), intermediate-lived (40–80 min), and long-lived (80+ min) groups. Long-lived cells were broader (∼13.2 km at 2–4-km height) and deeper (∼11.4 km) than short-lived cells (∼6.4-km width, ∼7.31-km height). Using random forest (RF) modeling and correlation analyses, precipitable water vapor (PWV), 2–6-km lapse rate, 0–8-km bulk shear, and fine aerosol mass concentration (Mass_f) are identified as key predictors of cell lifetime. Higher PWV is associated with significantly longer convective cell lifetimes compared to the low-PWV group, particularly within low 2–6-km temperature lapse rate (LR_26km), moderate-to-higher 0–8-km bulk shear (BS_08km), and low-to-moderate Mass_f environments. RF analysis also identifies low-level (0–2 km) equivalent potential temperature, PWV, Mass_f, and surface latent heat flux as key predictors for cell width and height. Short-lived cells have higher aerosol number concentrations (500–1000-nm size range), linked to onshore wind conditions and marine aerosols; however, their low concentration suggests the sensitivity may reflect associated meteorological regimes rather than a direct aerosol effect. Long-lived cells have higher concentrations of organic and sulfate aerosols, while short-lived cells exhibit higher black carbon concentrations. These results highlight the intricate dependence of convective cell lifetimes and structure on environmental moisture, thermodynamics, wind shear, and aerosol characteristics.

54 ENVIRONMENTAL SCIENCES↗
DOE OSTI2024

A deep learning-guided automated workflow in LipidOz for detailed characterization of fungal fatty acid unsaturation by ozonolysis

Understanding fungal lipid biology and metabolism is critical for antifungal target discovery as lipids play central roles in cellular processes. Nuances in lipid structural differences can significantly impact their functions, making it necessary to characterize lipids in detail to enable and understanding of their roles in these complex systems. In particular, lipid double bond (DB) locations are an important component of lipid structure that can only be determined using a few specialized analytical techniques. Ozone-induced dissociation mass spectrometry (OzID-MS) is one such technique that uses ozone to break lipid DBs, producing pairs of characteristic fragments that allow the determination of DB positions. In this work we apply OzID-MS and LipidOz software to analyze the complex lipids of Saccharomyces cerevisiae yeast strains transfected with different fatty acid desaturases from Histoplasma capsulatum to determine the specific unsaturated lipids produce. The automated data analysis in LipidOz made the determination of DB positions from this large dataset more practical, but manual verification for all targets was still time-consuming. The DL model reduces manual involvement in data analysis, but since it was trained using mammalian lipid extracts, the prediction accuracy on yeast-derived data was reduced. We addressed both shortcomings by retraining the DL model to act as a pre-filter to prioritize targets for automated analysis, providing confident manually verified results but requiring less computational time and manual effort. Our workflow resulted in the determination of novel DB positions and enzymatic specificity.

mass spectrometry, deep learning, Lipidomics, doub↗
DOE OSTI2024

Deep Learning-Based Dynamic Modeling of Three-Phase Voltage Source Inverters

Inverter-based resource (IBR) models are necessary to analyze modern power system stability and create effective control strategies. Modeling IBRs in converter-rich power systems is crucial, yet challenging due to the lack of commercial information on converter topologies and control parameters. This paper proposes novel convolutional neural network (CNN)–based data-driven techniques for modeling IBRs, addressing adaptability and proprietary concerns without requiring internal system physics knowledge. The proposed method is tested using real grid-tied commercial IBR transient data and demonstrates effectiveness and accuracy. Furthermore, the developed modeling approach is integrated and implemented in the open-source power distribution simulation and analysis tool, GridLAB-D, to illustrate the potentiality of dynamic analysis of large-scale power systems with high IBRs.

deep learning, artificial intelligence↗
DOE OSTI2025

Origin of proton irradiation-induced deep acceptors in Al 0.70 Ga 0.30 N

Deep level defect introduction and carrier removal were characterized using steady-state photocapacitance (SSPC), deep level transient spectroscopy (DLTS) and lighted capacitance–voltage for proton irradiated n-type Al 0.70 Ga 0.30 N Schottky diodes grown by metal-organic vapor phase epitaxy on AlN-on-sapphire templates. SSPC observed deep levels in the as-grown diode with zero-phonon transition energies of 2.20, 2.65, 3.10, 3.40, and 4.65 eV relative to the conduction band minimum (E c ), and an additional deep level emerged at 1.20 eV with irradiation. Lighted capacitance–voltage measurements quantified the deep level concentration (N t ) of states detected by SSPC, and it was observed that N t increased with proton fluence only for the 1.2 and 4.65 eV levels. Carrier removal was much larger than the increase in N t of the 1.2 and 4.65 eV deep levels, suggesting that radiation-induced deep level compensators existed beyond what was detected with SSPC. DLTS detected additional, proton-induced deep acceptors at E c —0.55, 0.82, and 1.16 eV, the latter of which is likely the same 1.20 eV deep state observed by SSPC. The concentration of the E c —0.82 eV defect state was large enough to reconcile carrier removal with total deep level introduction. Comparing the E c —0.82 and 1.16 eV deep acceptor levels to previous experimental and theoretical reports suggests that their atomistic origins could be the nitrogen vacancy (V N ) and oxygen substituting on the nitrogen sub-lattice (O N ), respectively. This defect behavior contrasts starkly with GaN, where V N and O N are shallow donors, and demonstrates that the electronic properties of defects can evolve drastically within the AlGaN alloy system ranging from wide bandgap GaN to ultra-wide bandgap AlN.

Armstrong, Andrew Michael [Sandia National Laborat↗
DOE OSTI2025

Systems Analysis of Biomass and Coal Co-firing Power Plants with Deep Carbon Capture Toward Net-zero Emissions

Achieving a net-zero emission economy in the United States requires integrating diverse low-carbon and negative-emission technologies into the existing fossil fuel-dominant power fleet. Potential technologies from the low-carbon portfolio include renewable power, fossil power with carbon capture and storage (CCS), bioenergy with CCS (BECCS), and direct air capture (DAC). Renewable power is a clean energy source but has to pair with costly battery storage to provide dispatchable electricity. Fossil power with CCS offers dispatchable electricity yet still relies on DAC to offset residual emissions, even when deploying deep CCS with more than 90% CO2 capture. Coal-biomass co-firing with CCS, a subset of BECCS, is a reliable energy production technology that can be retrofitted from existing electricity generation units (EGUs). Power plant retrofit maximizes the use of the current U.S. coal power fleet without the need for large-scale deployment of new renewable power, battery storage, or DAC. Retrofitting coal-biomass co-firing with deep CCS in EGUs is a promising option, but not a universal solution. Biomass co-firing at a power plant introduces economic challenges and indirectly poses pressure on land and water resources. Meanwhile, retrofitting deep CCS affects plant efficiency and raises electricity generation costs. Overall, the technical feasibility and economic viability of plant retrofits vary across EGUs, as they are contingent upon the regional availability of biomass, unit-specific characteristics, site-specific fuel supply costs, and adjacent CO2 storage potential. Government incentives like 45Q can improve the retrofit viability, though the impact requires further quantification. A comprehensive analysis at the unit level is essential to address the question regarding the fate of the U.S. coal-fired electricity generation fleet toward the net-zero emission goal. This study conducts a systematic techno-economic-environmental assessment of EGUs to identify the viability of biomass co-firing and deep CCS retrofits in the U.S. coal-fired power fleet. Specifically, it characterizes the techno-economic performance of deep carbon capture, estimates life cycle greenhouse gas (GHG) emissions, and conducts a fleet-level assessment on retrofit viability. The key objectives are (1) to estimate the unit-specific performance and retrofitted cost under various biomass co-firing levels and CO2 capture rates; (2) to determine the possibility of reaching net-zero emission at the fleet level; (3) to quantify the cumulative capacities that are suitable for plant retrofits under current and future biomass supply scenarios; and (4) to improve the understanding of policy impacts on such retrofits to help the power sector’s transition to a net-zero economy. Techno-economic Model of Deep Carbon Capture. This study develops the performance and economic models for Monoethanolamine-based post-combustion CO2 capture at 95–99% capture rates. The process is simulated in Aspen Plus, analyzing the performance of carbon capture technology by varying the plant sizes, solvent lean loading, CO2 concentrations, and flue gas inlet temperature. Based on the key inputs and output parameters of CO2 capture, a reduced-order performance model of deep carbon capture is formulated. In addition, an engineering-economic model integrating the performance metrics is developed to estimate the capital as well as operation and maintenance (O&M) costs. Capital cost estimations follow the framework of the Integrated Environmental Control Model (IECM) and incorporate data regressions from three technical reports by IECM, the National Energy Technology Laboratory (NETL), and the National Renewable Energy Laboratory. The O&M cost estimation utilizes the actual inventory consumption rate and labor requirements. Both performance and cost models are embedded into IECM v13.0-beta, a fossil-fuel power plant modeling tool. Life Cycle Assessment of Power Plants. This study estimates the GHG emissions of power plants through life cycle assessment (LCA). The LCA scope includes fuel supply, combustion-based power generation, and CO2 transport and storage. The fuel-based life cycle module is designed following the framework of the NETL Unit Process Library and CO2U LCA Guidance Toolkit. The module is then incorporated into IECM v13.0-beta. The process-based LCA is applied to estimate the GHG emissions of coal and biomass supply, coal- and coal-biomass co-firing power plant operation, as well as CO2 pipeline transport and geographical sequestration. An uncertainty analysis is conducted to quantify the variability and uncertainty associated with the LCA using the Latin Hypercube Sampling (LHS) method. Fleet-level Assessment. This study evaluates the technical and economic feasibility of selected coal-fired EGUs, examines the role of tax credits in retrofit viability, and assesses the competitiveness of retrofitted units against other low-carbon options. Unit screening identifies EGUs for the study, focusing on new, efficient baseload units with air pollution controls. The power plant databases are then established to organize unit-specific information on performance and operating conditions from the relevant public databases. Biomass for co-firing retrofits is selected based on home and neighboring county availability, ensuring sustained operation with at least a 5% co-firing level. The CO2 storage site is determined by state-level storage potential, with ArcGIS Pro and NETL CO2 Saline Storage Cost Model used to identify the optimal balance between the nearest transport distances and affordable storage costs. The latest IECM v13.0-beta is then employed to configure and evaluate the eligible EGUs with or without the deployment of deep CCS and biomass co-firing. A supply curve is established to illustrate the cumulative installed capacity suitable for retrofits at different cost levels. A sensitivity analysis on tax credits for carbon sequestration is performed. Finally, a unit-level cost comparison is conducted among retrofitted plants, renewable power with battery storage, and abated fossil fuels with DAC. Expected Results. This study evaluates the technical, economic, and environmental metrics of each EGU across an array of CO2 capture rates and biomass co-firing level scenarios. Unit-level comparisons will identify critical factors influencing technical performance. The supply curves with and without tax incentives will provide insights into the impact of tax credits on biomass co-firing and CCS deployment. The cost comparisons with renewables and DAC-retrofit will assess the competitiveness of the retrofitted units. Life cycle emissions from each unit will be assessed to identify the scenarios under which net-zero emissions can be achieved. These analyses are expected to determine the total coal-fired capacity suitable for serving as a low-carbon energy source with or without tax incentives. The study results are novel in identifying optimal unit-specific strategies for producing carbon-neutral power, whether through retrofitting EGUs with deep CCS, biomass co-firing, DAC, or installing renewable power with battery. The findings will provide insight into nationwide efforts to ensure reliable, affordable, and low-carbon electricity. It also will inform investment decisions and policies in the deployment of deep carbon capture and negative emission technologies for a net-zero energy future.

Biomass Co-firing↗
DOE OSTI2026

Examining Aerosol Vertical Transport and Removal during Deep Convective Events

Atmospheric aerosols affect the global energy budget by scattering and absorbing sunlight (direct effects) and by changing the microphysical structure, lifetime, and coverage of clouds (indirect effects). Globally, the free troposphere is a major source of nucleation- and Aitken-mode aerosols due to the enhanced new particle formation rates at high altitudes. Recent studies have shown deep convective systems are capable of transporting these small aerosols from the free troposphere to the boundary layer by strong convective downdrafts and weaker downward motions in the stratiform regions. These vertically transported aerosols can grow into cloud condensation nuclei (CCN) and play a significant role in the global climate. During the deep convective processes, existing accumulation-mode aerosols that act as coagulation sinks of smaller particles are also removed by wet scavenging. Compared to the vertical transport of these particles by entrainment mixing, which is slower but more prevalent, the deep convective downdraft processes may be more rapid and efficient in the vertical transport of aerosols. However, most of the current climate models do not include this mechanism as a source of CCN, mainly because the frequency of deep convective events varies significantly with geographic location and thus their contributions to CCN are unpredictable. We target this critical gap in understanding the vertical transport and removal of aerosols by deep convections. We proposed to analyze a multi-year, multi-site measurement record available from the U.S. Department of Energy (DOE) ARM program, including the observations from the 2014/15 Observations and Modeling of the Green Ocean Amazon (GoAmazon) field campaign, the 2017/18 Aerosol and Cloud Experiments in the Eastern North Atlantic (ACE-ENA) field campaign, the 2018/19 Cloud, Aerosol, and Complex Terrain Interactions (CACTI) field campaign, the 2021/22 Tracking Aerosol Convection Interactions Experiment (TRACER) field campaign, and the long-term measurements collected at the Southern Great Plains (SGP) atmospheric observatory, where deep convective clouds were frequently observed. This project is aimed at the following three objectives: (1) Gaining a detailed and quantitative understanding of the aerosols transported by a convective downdraft and their evolution in the atmosphere; (2) Examining the wet scavenging mechanisms and efficiencies of aerosols at altitudes of deep convective systems based on ground and aircraft measurements; (3) Evaluating the contribution of deep convective systems to CCN as both a source and a sink of atmospheric aerosols and its seasonal variabilities.

54 ENVIRONMENTAL SCIENCES↗
DOE OSTI2026

Deep-field analytical calibration

The next generation of imaging surveys, including the Vera C. Rubin Observatory Legacy Survey of Space and Time (LSST), Euclid, and the Nancy Grace Roman Space Telescope, will provide unprecedented constraints on cosmology using weak gravitational lensing. To fully exploit this statistical power, shear measurement methods must achieve sub- per cent accuracy while mitigating systematic biases from noise, the point-spread function (PSF), blending, and shear-dependent detection. The analytical calibration framework (AnaCal) has demonstrated such accuracy but requires adding noise to images, reducing effective depth. We introduce Deep-Field Analytical Calibration (DEEP-FIELD AnaCal), an extension of AnaCal that uses deep-field images to compute shear responses while preserving the statistical power of wide-field data. We validate DEEP-FIELD AnaCal on isolated and blended galaxy image simulations with LSST-like conditions, finding it meets the stringent requirement of multiplicative bias $|m| < 3\times 10^{-3}$ at 99.7 per cent confidence. Compared to standard AnaCal applied to wide-field images, DEEP-FIELD AnaCal increases the effective galaxy number density from 17 to 30 arcmin$^{-2}$ for simulated 10-yr LSST data. With deep fields $10\times$ longer than the wide field, we find pixel noise variance in shear estimation is reduced by 30 per cent and overall uncertainty by $\sim 25~{{\ \rm per\ cent}}$. Finally, using the LSST Deep Drilling Fields strategy, we assess sample variance and find an equivalent calibration uncertainty of $\lesssim 0.3~{{\ \rm per\ cent}}$. These results demonstrate that DEEP-FIELD AnaCal offers a promising path to achieve the required shear calibration for upcoming weak lensing surveys.

79 ASTRONOMY AND ASTROPHYSICS↗
DOE OSTI2024

Evidential Deep Learning: Enhancing Predictive Uncertainty Estimation for Earth System Science Applications

Abstract Robust quantification of predictive uncertainty is a critical addition needed for machine learning applied to weather and climate problems to improve the understanding of what is driving prediction sensitivity. Ensembles of machine learning models provide predictive uncertainty estimates in a conceptually simple way but require multiple models for training and prediction, increasing computational cost and latency. Parametric deep learning can estimate uncertainty with one model by predicting the parameters of a probability distribution but does not account for epistemic uncertainty. Evidential deep learning, a technique that extends parametric deep learning to higher-order distributions, can account for both aleatoric and epistemic uncertainties with one model. This study compares the uncertainty derived from evidential neural networks to that obtained from ensembles. Through applications of the classification of winter precipitation type and regression of surface-layer fluxes, we show evidential deep learning models attaining predictive accuracy rivaling standard methods while robustly quantifying both sources of uncertainty. We evaluate the uncertainty in terms of how well the predictions are calibrated and how well the uncertainty correlates with prediction error. Analyses of uncertainty in the context of the inputs reveal sensitivities to underlying meteorological processes, facilitating interpretation of the models. The conceptual simplicity, interpretability, and computational efficiency of evidential neural networks make them highly extensible, offering a promising approach for reliable and practical uncertainty quantification in Earth system science modeling. To encourage broader adoption of evidential deep learning, we have developed a new Python package, Machine Integration and Learning for Earth Systems (MILES) group Generalized Uncertainty for Earth System Science (GUESS) (MILES-GUESS) ( https://github.com/ai2es/miles-guess ), that enables users to train and evaluate both evidential and ensemble deep learning. Significance Statement This study demonstrates a new technique, evidential deep learning, for robust and computationally efficient uncertainty quantification in modeling the Earth system. The method integrates probabilistic principles into deep neural networks, enabling the estimation of both aleatoric uncertainty from noisy data and epistemic uncertainty from model limitations using a single model. Our analyses reveal how decomposing these uncertainties provides valuable insights into reliability, accuracy, and model shortcomings. We show that the approach can rival standard methods in classification and regression tasks within atmospheric science while offering practical advantages such as computational efficiency. With further advances, evidential networks have the potential to enhance risk assessment and decision-making across meteorology by improving uncertainty quantification, a longstanding challenge. This work establishes a strong foundation and motivation for the broader adoption of evidential learning, where properly quantifying uncertainties is critical yet lacking.

Schreck, John S.↗
DOE OSTI2025

Earthquake-triggered submarine canyon flushing transfers young terrestrial and marine organic carbon into the deep sea

Submarine canyons transfer substantial amounts of sediment and organic carbon (OC) into the deep ocean, nourishing deep-sea ecosystems and contributing to the global carbon cycle through OC burial and sequestration. Tracking lateral OC transport through submarine canyon systems is challenged by the deep-ocean setting, difficulties with constraining episodic depositional events, and the need to assess the composition and age of marine and terrestrial organic matter. We apply innovative parallel ramped pyrolysis oxidation accelerator mass spectrometry and pyrolysis-gas chromatography-mass spectrometry with isotope analyses to track OC age and sources in the 2016 Kaikoura earthquake-triggered, canyon-flushing event that deposited along >1300 km of a submarine canyon-channel system, offshore Aotearoa New Zealand. Specifically, these techniques allow us to determine the ages, sources, and partitioning of OC within the Kaikoura turbidite deposit and test hypotheses of how submarine canyon systems contribute to lateral OC flux and burial. Our results show that, despite considerable canyon floor erosion, substantial amounts of young OC were flushed into the deep sea, with relatively little (~2%) pre-Holocene OC contributions. Even without a direct connection between rivers and submarine canyons, most (~55%) of the OC in the Kaikoura event bed is from terrestrial sources. However, the deposit also contains substantial amounts (~22%) of marine-derived OC and ~23% of the material is of unassignable origin. Particle sorting imparts variability on the age and composition of OC within turbidite deposits and along the turbidity current flow path. Terrestrial-derived OC is preferentially older than marine-derived OC and concentrated in coarser particle sizes found more commonly at the deposit base and in proximal settings. Young, marine-derived OC is concentrated at the surface of the deposits and tends to be enriched in finer particle sizes. Such OC partitioning in turbidites supports the relevance of depositional models for predicting and quantifying distribution of OC in deep-sea deposits. Earthquake-triggered, canyon flushing events and resulting turbidites enhance OC burial efficiency and can sequester OC effectively, contributing an important carbon sink to the sedimentary carbon cycle.

54 ENVIRONMENTAL SCIENCES↗
DOE OSTI2025

Sensitivity of the Shallow‐To‐Deep Convective Transition to Moisture and Wind Shear in the Amazon

Deep convection is the primary influence on weather and climate in tropical regions. However, understanding and simulating the shallow‐to‐deep (STD) convective transition has long been challenging. Here, we conduct high‐resolution numerical simulations to assess the environmental controls on the evolution of isolated convection in the Amazon during the wet season. The large‐scale forcing derived through a constrained variational analysis approach for the GoAmazon2014/5 Experiment is used in the simulations. Through sensitivity experiments, we examine the relative importance of moisture and wind shear in controlling the shallow‐to‐deep convective transition for isolated convective events. Convection exhibits the greatest sensitivity to humidity within the lowest 1.5 km, where a 4 mm reduction in column water vapor nearly suppresses ice water formation on deep convective days. In contrast, a reduction in column water vapor in the free troposphere by a factor of two or more is necessary to produce a comparable impact on convection. Increasing low‐level wind speed from 6 to 9 m s -1 enhances afternoon deep convection, raising the cloud ice mixing ratio by approximately 25%. Conversely, upper‐level wind shear reveals the weakest correlation with daytime convection in our simulations. Our results help characterize the role of moisture and wind shear on the STD transition and our understanding of the underlying mechanisms.

environmental controls↗
DOE OSTI2026

Deep Neural Networks are Adaptive to Function Regularity and Data Distribution in Approximation and Estimation

Deep learning has exhibited remarkable results across diverse areas. To understand its success, substantial research has been directed towards its theoretical foundations. Nev- ertheless, the majority of these studies examine how well deep neural networks can model functions with uniform regularities. In this paper, we explore a different angle: how deep neural networks can adapt to varying degrees of smoothness in functions and nonuni- form data distributions across different locations and scales. More precisely, we focus on a broad class of functions defined by nonlinear tree-based approximation methods. This class encompasses a range of function types, such as functions with uniform regularities and discontinuous functions. We develop nonparametric approximation and estimation theories for this class using deep ReLU networks. Our results show that deep neural networks are adaptive to the nonuniform smoothness of functions and nonuniform data distributions at different locations and scales. We apply our results to several function classes, and derive the corresponding approximation and generalization errors. The validity of our results is demonstrated through numerical experiments.

97 MATHEMATICS AND COMPUTING↗