Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “Distribution System State Estimation”

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 793 records · Page 44

Cyberattack Detection and Mitigation on Central Volt‐VAr Using Circuit Law and Machine Learning

ABSTRACT In a distribution grid, voltage is maintained within a nominal range through a Volt‐VAr function that controls capacitor banks, reactive power of distributed energy resources (DER), and on‐load tap changers (OLTC). Availability of communications helps with the implementation of central Volt‐VAr control; however, it also opens the system to cyberattacks, causing voltage disturbances. Previous work has shown the adverse impacts of false data injection (FDI) on the central Volt‐VAr control; however, very few works have studied methods to detect and mitigate FDI on Volt‐VAr control. This paper addresses gaps in the detection and mitigation of FDI on the measurement packets of a central Volt‐VAr control. This work uses a two‐stage algorithm for cyberattack detection since the accuracy of a single‐stage machine learning (ML)–based detection method decreases while dealing with unseen data. The first stage is based on the verification of measurements against circuit laws, and the second stage utilizes a tree search algorithm and an ML method to detect the falsified data. This paper compares long short‐term memory (LSTM) and bidirectional LSTM (BiLSTM) as the employed ML algorithms. Finally, the mitigation algorithm replaces the falsified data with the estimated output of the ML algorithm. The effectiveness of the proposed method is tested for several cases using the IEEE 13‐bus test system in PSCAD software.

Beikbabaei, Milad [Bradley Department of Electrica↗

Electronic structures of [001]- and [111]-oriented InSb and GaSb free-standing nanowires

We report on a theoretical study of the electronic structures of InSb and GaSb nanowires oriented along the [001] and [111] crystallographic directions. The nanowires are described by atomistic, tight-binding models, including spin-orbit interaction. The band structures and the wave functions of the nanowires are calculated by means of a Lanczos iteration algorithm. For the [001]-oriented InSb and GaSb nanowires, the systems with both square and rectangular cross sections are considered. Here, it is found that all the energy bands are doubly degenerate. Although the lowest conduction bands in these nanowires show good parabolic dispersions, the top valence bands show rich and complex structures. In particular, the topmost valence bands of the nanowires with a square cross section show a double maximum structure. In the nanowires with a rectangular cross section, this double maximum structure is suppressed, and the top valence bands gradually develop into parabolic bands as the aspect ratio of the cross section is increased. For the [111]-oriented InSb and GaSb nanowires, the systems with hexagonal cross sections are considered. It is found that all the bands at the Γ-point are again doubly degenerate. However, some of them will split into non-degenerate bands when the wave vector moves away from the Γ-point. Although the lowest conduction bands again show good parabolic dispersions, the topmost valence bands do not show the double maximum structure. Instead, they show a single maximum structure with its maximum at a wave vector slightly away from the Γ-point. The wave functions of the band states near the band gaps of the [001]- and [111]-oriented InSb and GaSb nanowires are also calculated and are presented in terms of probability distributions in the cross sections. It is found that although the probability distributions of the band states in the [001]-oriented nanowires with a rectangular cross section could be qualitatively described by one-band effective mass theory, the probability distributions of the band states in the [001]-oriented nanowires with a square cross section and the [111]-oriented nanowires with a hexagonal cross section show characteristic patterns with symmetries closely related to the irreducible representations of the relevant double point groups and, in general, go beyond the prediction of a simple one-band effective mass theory. We also investigate the effects of quantum confinement on the band structures of the [001]- and [111]-oriented InSb and GaSb nanowires and present an empirical formula for the description of quantization energies of the band edge states in the nanowires, which could be used to estimate the enhancement of the band gaps of the nanowires as a result of quantum confinement. The size dependencies of the electron and hole effective masses in these nanowires are also investigated and discussed.

71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSIC↗

NFD Methodology Report

The NetFlow Dynamics (NFD) model was developed for estimating the availability of a commodity supplied by a national- or regional-scale infrastructure following unexpected disruption of one or more of its components. The large scope of the disruptions of interest produce changes in availability lasting days to weeks. Consequently, the model does not resolve daily variations in system state and does not include the long-term processes that cause infrastructures to evolve as assets are added and removed according to owners ’planning decisions. NFD simulates fluid flow, including petroleum and other incompressible fluids, as well as natural gas and other compressible fluids, through pipeline networks characterized by limits on transmission capacity and storage. It was designed to enable efficient exploration of possible transmission system responses to large-scale disruptions lasting for days or longer. The model formulation reflects constraints on transmission and storage capacity imposed by the physical system assets. Those capacity limits are input parameters and are not derived from more basic system properties such as pipeline diameters and compressor power. A system’s response to a large disruption is controlled by operational decisions as well as damage to physical assets. The NFD model formulation allows users to efficiently consider alternative scenarios about the way remaining capacity might be used so that the analysis result appropriately reflects uncertainties about operator response.

24 POWER TRANSMISSION AND DISTRIBUTION↗

Probabilistic Approach to Quantifying the Contribution of Variable Generation and Transmission to System Reliability: Preprint

The increasing electrical load served by variable generation (VG), such as wind and solar energy, in the United States and many other countries has stimulated an interesting line of research to better quantify the capacity value of these resources. Methods applied traditionally to thermal units based on their average outage rates do not apply to VG because of their uncertain and non-dispatchablenature. The North American Electric Reliability Corporation's Integration of Variable Generation Task Force recently released a report that highlighted the need to develop and benchmark underlying loss-of-load expectation and related metrics that reasonably and fairly calculate the contribution to planning reserves, or capacity value, of solar and wind power. As the fraction of generation comingfrom VG becomes more significant, their estimated capacity value will have a larger impact on system planning. In this paper, we provide a method to include VG in traditional probabilistic-based adequacy methods. This method has been implemented in the Renewable Energy Probabilistic Resource Assessment tool developed at the National Renewable Energy Laboratory. Through an example based on the U.S. Western Interconnection, this method is applied to assess the effect that transmission can have on resource adequacy. We also analyze the interactions between available transmission and capacity value for VG.

14 SOLAR ENERGY↗

Kernel Density Estimators for SAMPLER Analyses [Slides]

This presentation discusses Sampler, specifically in terms of uncertainty quantification and states that for many simulation tasks nominal input parameters have complicated influence on the uncertainty in the final result. Sampler provides a convenient means to perform stochastic uncertainty quantification, brute force sampling parameter space and tallying changes in outputs, and it can be coupled with any SCALE modules. Perturbed responses can be collected and analyzed with standard statistical measures; however, two problems have been identified. The first problem is that for Monte Carlo simulations the individual response uncertainties are ignored, and the second problem is that features of distributions are easily hidden in histograms. They found that the solution is to add Kernel Density Estimator (KDE) analysis to the Sampler output. In conclusion, Sampler is a unique tool for stochastic uncertainty propagation, and it can be coupled with any module in SCALE. Current functionality is limited to histograms, and it may not adequately account for sample uncertainties and/or not adequately display underlying distribution features. Work is in progress to implement KDE in the Sampler module and no additional flagging is needed.

73 NUCLEAR PHYSICS AND RADIATION PHYSICS↗

Visible spectroscopy diagnostics for tungsten source assessment in the WEST tokamak: First measurements

The present work concerns the measurements obtained with the Tungsten (W) Environment in Steady-state Tokamak (WEST) visible spectroscopy system during the first experimental campaign. This system has been developed in the framework of the WEST project that equipped the existing Tore Supra device with a tungsten divertor in order to test actively cooled tungsten Plasma Facing Components (PFC) in view of preparing for ITER operation. The goal of this diagnostic is to measure the PFC sources and the deuterium recycling with spectral, spatial, and temporal resolution adapted to the predicted power deposition profiles on the objects observed. Three kinds of PFCs are monitored: the Ion Cyclotron Resonance Heating (ICRH) antenna and Low Hybrid Current Drive (LHCD) launcher W limiters; one of the 6 W inner bumpers; and the upper and lower W divertors. Large-aperture in-vessel actively cooled optical systems (f-number ~ 3) were installed for each view and connected to optical fibres. A total of 240 optical fibers can be distributed on various detection systems including a fast response-time, multi-channel, filtered photodetector-based “Filterscope” system, developed by Oak Ridge National Laboratory (USA) as well as grating spectrometers optimized for multi-sightline analysis. The first WEST experimental campaign conducted in 2017 has been dedicated to plasma start-up development during which the visible spectroscopy system has provided crucial information related to the impurity content first and then impurity sources. The diagnostic setup for that first experimental campaign was limited to the inner bumper and outer limiters but was sufficient to demonstrate that the optical setup was in accordance with the specifications. The radiance calibration procedure allowed us to estimate fluxes from the main limiter of about 8 × 10 18 atoms/(s m 2 ) and to show a first W source radial profile along the outboard limiter.

Meyer, O.↗

Towards a Robust Adaptive Digital Twin for Fusion Applications

The development of a digital twin system for fusion applications is essential for enhancing the prediction, analysis, and optimization of complex plasma processes. Machine learning (ML), particularly deep learning has demonstrated strong capabilities in modeling such highly nonlinear and intricate systems. However, two critical challenges limit the deployment of deep learning-based digital twins: Uncertainty Quantification (UQ) and data drift. UQ is vital for ensuring trustworthy predictions, especially in decision-support scenarios. Additionally, data-driven models are often sensitive to changes in the underlying data distribution, such as shot-to-shot variations in fusion experiments, which can lead to performance degradation over time. To address these challenges, we are developing an uncertainty-aware, adaptive digital twin framework. Our approach incorporates deep learning models enhanced with Gaussian Process approximations for predictive uncertainty estimation, coupled with an online learning mechanism that enables continuous model adaptation to new experimental data. This adaptive capability allows the data driven models to respond effectively to evolving plasma behaviors and equipment conditions. Specifically, to mitigate the effects of shot-to-shot drift, our system updates itself incrementally as new data becomes available, improving both robustness and fidelity. Our vision is to evolve this data driven model into a self-sustaining digital twin system that leverages UQ based feedback to continuously refine itself and potentially support real-time decision making. This presentation will cover a brief background on uncertainty quantification for ML, our ongoing effort on development of UQ capabilities for ML, our data science pipeline from data collection to model development and analysis and online learning framework for modeling coil deflection at DIII-D. I will also briefly touch upon opportunities and challenges in development of digital twin framework.

Sammuli, Brian [General Atomics]↗

Predicting Large‐Scale Systematic Missing Pipe Attributes in Water Distribution Networks

Water distribution network (WDN) models are an essential tool used by water utilities for hydraulic analysis. Unfortunately, missing data and insufficient resources often make creating and maintaining these models unfeasible. Existing methods to address missing pipe properties, like sequential imputation for missing values and reconstruction using graph metrics, are designed to accommodate random patterns of missing information and require a significant percentage of the system's attributes to be known. However, these data completeness assumptions do not always align with real‐world scenarios where large sections of the WDN model have missing data. To address this challenge, this study proposes a data‐driven approach for estimating pipe diameter when considering different spatial patterns and degrees of data completeness (i.e., 0%–90%). Using data from 16 WDNs in Kentucky, this study compares the use of machine learning (ML) using topological and geospatial features against an existing deterministic approach. Results demonstrate that WDN models with pipe diameters predicted by the proposed ML method had comparable hydraulic performance to the ground truth models. Moreover, results showed that ML method performance varies between WDNs of differing topological classification. Insights from this study help advance the ability to leverage partial data to create and maintain WDN models amid uncertainty and inadequate resources.

Poff, Jason W. [Oregon State Univ., Corvallis, OR ↗

Analysis of Unpolarised p+p¿ Photoproduction with the GlueX Experiment

This thesis presents measurements of spin-density matrix elements in unpolarised p+p? photoproduction on a proton target. The dominant resonance contribution to the dipion system is the r(770) meson. Due to the large production cross section for this resonance, a valid comparison can be made between the obtained final results and r(770) spin-density matrix elements measured previously with other experiments. The measurement was performed over the 3:0 ? 11:6 GeV beam energy regime, which is a more extensive energy range than has ever been studied previously for the r(770). Results were obtained by analysing data from the GlueX experiment based at Jefferson Lab. Extended maximum likelihood fits were applied to extract three spin-density matrix elements using Markov chain Monte Carlo based parameter estimations. This was performed using various binning configurations to probe the energy, mass, and four-momentum transfer dependence of the determined physics observables. Spin-density matrix elements are shown to be consistent with the model of s-channel helicity conservation at low ?t. The effects of pomeron and f2 exchanges are clearly visible in the energy dependence of the measured observables. These observations provide valuable insights into the relative strengths of both processes as a function of the photon energy, and may enable theorists to disentangle the f2=P coupling ratio. Spin-density matrix elements are seen to be highly dependent on the reconstructed resonance mass. This observation is likely to be a result of non-resonant S-wave background processes, and emphasises the need for a more detailed model of the p+p? angular distribution that considers all of the competing angular momentum components that contribute to the measured final state. The statistical precision of measurements performed for this thesis surpass what was achievable in previous studies of the r(770) by several orders of magnitude. Studies of the energy and four-momentum transfer dependence, and insights into the effects of non r(770) background contributions provide valuable input for production models. This will help inform the choice of wave-sets used for partial wave analyses, supporting GlueX in its search for exotic hybrid meson states.

Fitches, James↗

Phase Identification in Real Distribution Networks with High PV Penetration Using Advanced Metering Infrastructure Data

Many distribution network monitoring and control applications - including state estimation, volt/VAR optimization, and network reconfiguration - rely on accurate network models; however, the network models maintained by utilities can become outdated because of restoration activities, network reconfiguration, and missing data. With the widespread deployment of advanced metering infrastructure (AMI), abundant measurement data from low-voltage secondary networks are available. The AMI measurement data can be used for phase identification to improve the network models. Although the existing phase identification techniques work well in passive distribution feeders that do not have photovoltaic (PV) generation, they can fail to accurately identify the phases in the presence of PV. This paper proposes a robust phase identification algorithm based on supervised machine learning that accurately identifies the AMI meter phase connectivity in the presence of significant PV generation. The proposed algorithm does not require network topology information or feeder head measurement data. The algorithm is validated using the AMI measurement data collected in the field and the field-validated phase connectivity database on two real distribution feeders from San Diego Gas & Electric Company that have significant PV generation.

advanced metering infrastructure↗

Phase Identification in Real Distribution Networks with High PV Penetration Using Advanced Metering Infrastructure Data: Preprint

Many distribution network monitoring and control applications - including state estimation, volt/VAR optimization, and network reconfiguration - rely on accurate network models; however, the network models maintained by utilities can become outdated because of restoration activities, network reconfiguration, and missing data. With the widespread deployment of advanced metering infrastructure (AMI), abundant measurement data from low-voltage secondary networks are available. The AMI measurement data can be used for phase identification to improve the network models. Although the existing phase identification techniques work well in passive distribution feeders that do not have photovoltaic (PV) generation, they can fail to accurately identify the phases in the presence of PV. This paper proposes a robust phase identification algorithm based on supervised machine learning that accurately identifies the AMI meter phase connectivity in the presence of significant PV generation. The proposed algorithm does not require network topology information or feeder head measurement data. The algorithm is validated using the AMI measurement data collected in the field and the field-validated phase connectivity database on two real distribution feeders from San Diego Gas & Electric Company that have significant PV generation.

advanced metering infrastructure↗

Cropland carbon fluxes in the United States: increasing Geospatial Resolution of Inventory-Based Carbon Accounting

Net annual soil carbon change, fossil fuel emissions from cropland production, and cropland net primary productivity were estimated and spatially distributed using land cover defined by the Moderate Resolution Imaging Spectroradiometer (MODIS) and by the Cropland Data Layer (CDL). Spatially resolved estimates of net ecosystem exchange (NEE) and net ecosystem carbon balance (NECB) were developed. NEE represents net on-site vertical fluxes of carbon. NECB represents all on-site and off-site carbon fluxes associated with crop production. Estimates of cropland NEE using moderate resolution (~1km2) land cover data were generated for the conterminous US and compared with higher resolution (30m) estimates of NEE and with direct measurements of CO2 flux from croplands in Illinois and Nebraska. Estimates of NEE using the CDL (30m resolution) had a higher correlation with eddy covariance flux tower estimates compared with estimates of NEE using MODIS. Estimates of NECB are primarily driven by net soil carbon change, fossil-fuel emissions associated with crop production, and CO2 emissions from the application of agricultural lime. NEE and NECB for US croplands were -274 and 7 Tg C yr-1 for 2004, respectively. Use of moderate to high resolution satellite-based land cover data enables improved estimates of cropland carbon dynamics.

29 ENERGY PLANNING, POLICY, AND ECONOMY↗

Benchmarking thermal energy storage cost for industrial process heat

Process heat accounts for roughly half of industrial energy demand, and currently 95% of process heat is derived from the combustion of natural gas, oil, and coal. Electrification of industrial heating could be an alternative, potentially expanding locations suitable for manufacturing; however, industrial facility owners may desire energy storage to stabilize energy costs. In this work, the economic benefits of pairing thermal storage with electrified process heat to reduce the average price paid for energy are analyzed. Cost savings focus on energy arbitrage, or leveraging flexible energy pricing schemes, alone. The cost of natural gas combustion across decades (2019-2060) is compared to the costs of electricity and thermal energy storage in four United States Independent System Operator (ISO) regions. Systems installed today may not yield positive net present value (NPV) compared to the use of natural gas. However, using estimated electricity prices, systems installed in 2030 using arbitrage alone could be profitable when compared to natural gas in some regions of the U.S. Furthermore, if capital expenditures could be reduced by 50% for sensible thermal storage systems by 2030, profitable systems are found across all regions. This implies that electrification of industrial process heat, when paired with inexpensive thermal energy storage systems, could be less expensive than brownfield natural gas systems, using arbitrage as the only source of revenue and without a dependency on any future policy drivers such as pricing externalities that could further incentivize the electrification of industrial process heat.

24 POWER TRANSMISSION AND DISTRIBUTION↗

An Integrated Modeling Framework for Sediment Dynamics During Urban Flooding: Application to Hurricane Harvey in Houston

Floodwater can mobilize and redistribute large volumes of sediment from upland to downstream urban areas, threatening infrastructure, water quality, and ecosystem health. However, existing modeling approaches often fail to capture sediment dynamics in urban floodplains due to the lack of integration between upland hydrological processes and riverine sediment transport. This study presents the first integrated modeling framework that couples the Energy Exascale Earth System Model (E3SM) land component, which simulates runoff and hillslope erosion, with TELEMAC-GAIA, a two-dimensional hydrodynamic and sediment transport model. This framework enables the fully distributed, process-based simulation of high-resolution (as fine as 30 m) sediment dynamics from hillslopes to floodplains. Applied to a highly urbanized watershed in Houston during Hurricane Harvey, this framework reproduced observed water levels at 16 USGS gauges (median R 2 = 0.83 and KGE = 0.78), key sediment dynamics such as sediment transport and deposition processes, and reproduced spatial deposition patterns consistent with LiDAR-derived data. Based on the simulation, we estimate 8.0 million m 3 of event-scale sediment deposition, including 5.7 million m 3 trapped in the flood-control reservoirs and 2.3 million m 3 deposited along major channels and floodplains. Using a representative unit removal cost, this corresponds to an estimated dredging cost of $581 million for total deposition. These results provide a first-order, physically based quantification of Harvey-scale sediment impacts. This study provides a valuable tool for the holistic analysis of sediment dynamics triggered by extreme urban flooding, supporting flood-resilience planning. More broadly, it highlights the importance of integrating physically based hydrological processes for urban flooding and sediment research.

Hurricane Harvey↗

Automatic Calibration and Health Monitoring of Infrastructure Sensors

Smart transportation infrastructure relies on networks of heterogeneous sensors - cameras, radars, and lidars - continuously monitoring traffic conditions. However, executing the initial spatial calibration of multiple sensors and the subsequent health monitoring presents significant operational challenges. Environmental factors, mechanical vibrations, and gradual drift cause spatial misalignment, degrading fusion performance and tracking accuracy. Traditional calibration approaches require manual intervention with specialized targets or survey equipment, resulting in service interruptions and high maintenance costs. This work presents an automated framework for initial calibration and continuous health monitoring without human intervention or service disruption. Our approach addresses two critical problems: (1) detecting when sensors become miscalibrated during operation, and (2) automatically re-establishing spatial alignment using only operational traffic data. The health monitoring component analyzes measurement innovations - differences between sensor observations and predicted object states - to detect systematic biases indicative of calibration drift. By computing bias magnitude, directional consistency, and rejection rates, the system identifies miscalibrations as small as 0.5 meters. Unlike traditional methods requiring known calibration targets, our diagnostic operates continuously on live traffic observations, enabling early detection before fusion quality degrades. The automatic recalibration algorithm leverages overlapping sensor fields-of-view and temporal correlation of vehicle observations. Using graph-based optimization, the system automatically discovers which sensor pairs observe common regions, estimates pairwise spatial transformations using RANSAC-based robust estimation, and jointly optimizes all sensor poses through bundle adjustment. The framework handles practical deployment challenges, including different sensor sampling rates (1-10 Hz), varying installation positions, unknown orientations, and limited overlap regions (>10%). When approximate sensor positions are available from installation surveys (+/-1m accuracy), the algorithm additionally estimates sensor orientations, refining both position and rotation to sub-meter and sub-degree accuracy. We validate the framework on multi-hour traffic datasets from six heterogeneous sensors with sampling rates ranging from 1 Hz to 10 Hz. Results demonstrate successful calibration even with sparse overlap (<20%) and automatic detection of miscalibrations exceeding 0.8 meters. This work enables a "deploy-and-forget" sensor infrastructure that maintains calibration autonomously, reducing maintenance costs while improving tracking accuracy. The techniques generalize beyond transportation to any multi-sensor monitoring application requiring robust spatial alignment, including smart cities, industrial monitoring, and surveillance systems.

24 POWER TRANSMISSION AND DISTRIBUTION↗

Residential Vehicle-to-Home Backup Power Capabilities: Key Findings from a ComEd Beneficial Electrification R&D Pilot

This report summarizes key findings from a collaborative technical study of residential, non-grid-tied vehicle-to-home (V2H) backup power systems in Commonwealth Edison’s (ComEd’s) service territory. The work integrates (1) a feeder-level technoeconomic analysis (TEA) using historical outage-event data and simulated electric-vehicle (EV) driving/charging profiles to estimate potential reliability and customer interruption-cost impacts under V2H and vehicle-to-grid (V2G) adoption scenarios; (2) controlled laboratory performance testing of a representative V2H backup ecosystem to characterize transfer-to-backup behavior, sustained power delivery, efficiency trends, and repeatable reliability limitations; and (3) a cybersecurity assessment aligned with NIST Cybersecurity Framework (CSF) 2.0 and ISO/SAE 21434 to evaluate interface-level risk drivers and identify program-relevant mitigations. Results indicate that V2H can provide measurable resilience value, but outcomes are strongly context dependent on outage patterns and the share of events that are “V2H-applicable.” Typical transfer-to-backup behavior clustered on the order of minutes, but rare long-delay edge cases were observed (including an event approaching 30 minutes) and should be treated as a reliability risk. High-power testing showed that peak-rated output is not necessarily continuously deliverable; stable operation may require operation below nameplate ratings and attention to thermal and installation constraints. The cybersecurity assessment highlights a broad attack surface spanning commissioning, home networks, embedded services, and cloud/OTA pathways, motivating minimum controls for secure onboarding, signed updates, patch cadence, and coordinated vulnerability response for any scaled deployment.

24 POWER TRANSMISSION AND DISTRIBUTION↗

Energy, power, and infrastructure demands from electrifying airport ground support equipment at United States airports

As the airline industry seeks to reduce costs and transition to clean energy, electric ground support equipment is emerging as a favorable option. The integration of electric ground support equipment into airport operations requires careful planning for vehicle deployment, charging infrastructure, and grid impacts. We develop a flexible, bottom-up modeling framework to assess energy and infrastructure needs across more than 300 U.S. airports. Our analysis estimates site-specific power and energy demand, equipment counts by type, charger requirements, and costs. Here we quantify the magnitude of new electrical loads created by the electrification of airport ground support equipment, finding that peak power demand at the largest airports can reach up to 20 megawatts, with annual electricity consumption approaching 51,000 megawatt-hours. We further show that behind-the-meter battery energy storage systems and solar photovoltaic systems can reduce peak load and lower total system costs by as much as 10 million dollars.

24 POWER TRANSMISSION AND DISTRIBUTION↗

Applying Gaussian Process Machine Learning and Modern Probabilistic Programming to Satellite Data to Infer CO 2 Emissions

Satellite data provides essential insights into the spatiotemporal distribution of CO 2 concentrations. However, many atmospheric inverse models fail to adequately incorporate the spatial and temporal correlations inherent in satellite observations and often lack rigorous methods for estimating parameters like spatial length scales. We introduce an inference model that processes the spatiotemporal covariance in satellite data and estimates hyperparameters such as covariance length scales. Our approach uses the Gaussian process (GP) machine learning (ML) and modern probabilistic programming languages (PPLs) to perform atmospheric inversions of emissions from satellite data. We develop a GP ML inversion system based on modern PPLs and the GEOS-Chem chemical transport model, simulating atmospheric CO 2 concentrations corresponding to the Orbiting Carbon Observatory-2/3 (OCO-2/3) data for July 2020. In our supervised learning framework, we treat the GEOS-Chem simulated data set as the target, with predictors derived by scaling the target with sector-specific factors hidden from the GP machine. Our results show that the GP model, combined with GPU-enabled PPLs, effectively retrieves true emission scaling factors and infers noise levels concealed within the data. This suggests that our method could be applied over larger areas with more complex covariance structures, enabling comprehensive analysis of the spatiotemporal patterns observed in OCO-2/3 and similar satellite data sets.

54 ENVIRONMENTAL SCIENCES↗