Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “Forward Modeling”

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 1,045 records · Page 58

Support Vector Machine-Based Prediction of Local Tumor Control After Stereotactic Body Radiation Therapy for Early-Stage Non-Small Cell Lung Cancer

Background: Several prognostic factors for local tumor control probability (TCP) after stereotactic body radiation therapy (SBRT) for early stage non-small cell lung cancer (NSCLC) have been described, but no attempts have been undertaken to explore whether a nonlinear combination of potential factors might synergistically improve the prediction of local control. Methods and Materials: We investigated a support vector machine (SVM) for predicting TCP in a cohort of 399 patients treated at 13 German and Austrian institutions. Among 7 potential input features for the SVM we selected those most important on the basis of forward feature selection, thereby evaluating classifier performance by using 10-fold cross-validation and computing the area under the ROC curve (AUC). The final SVM classifier was built by repeating the feature selection 10 times with different splitting of the data for cross-validation and finally choosing only those features that were selected at least 5 out of 10 times. It was compared with a multivariate logistic model that was built by forward feature selection. Results: Local failure occurred in 12% of patients. Biologically effective dose (BED) at the isocenter (BED{sub ISO}) was the strongest predictor of TCP in the logistic model and also the most frequently selected input feature for the SVM. A bivariate logistic function of BED{sub ISO} and the pulmonary function indicator forced expiratory volume in 1 second (FEV1) yielded the best description of the data but resulted in a significantly smaller AUC than the final SVM classifier with the input features BED{sub ISO}, age, baseline Karnofsky index, and FEV1 (0.696 ± 0.040 vs 0.789 ± 0.001, P<.03). The final SVM resulted in sensitivity and specificity of 67.0% ± 0.5% and 78.7% ± 0.3%, respectively. Conclusions: These results confirm that machine learning techniques like SVMs can be successfully applied to predict treatment outcome after SBRT. Improvements over traditional TCP modeling are expected through a nonlinear combination of multiple features, eventually helping in the task of personalized treatment planning.

62 RADIOLOGY AND NUCLEAR MEDICINE↗

Hydride-based thermal energy storage

Abstract The potential and research surrounding metal hydride (MH) based thermal energy storage is discussed, focusing on next generation thermo-chemical energy storage (TCES) for concentrated solar power. The site availability model to represent the reaction mechanisms of both the forward and backward MH reaction is presented, where this model is extrapolated to a small pilot scale reactor, detailing how a TCES could function/operate in a real-world setting using a conventional shell & tube reactor approach. Further, the important parameter of effective thermal conductivity is explored using an innovative multi-scale model, to providing extensive and relevant experimental data useful for reactor and system design. Promising high temperature MH material configurations may be tuned by either destabilisation, such as using additions to Ca and Sr based hydrides, or by stabilisation, such as fluorine addition to NaH, MgH 2 , or NaMgH 3 . This versatile thermodynamic tuning is discussed, including the challenges in accurately measuring the material characteristics at elevated temperatures (500 –700 °C). Attention to scale up is explored, including generic design and prototype considerations, and an example of a novel pilot-scale pillow-plate reactor currently in development; where materials used are discussed, overall tank design scope and system integration.

25 ENERGY STORAGE↗

Apparent dust size discrepancy in aerosol reanalysis in north African dust after long-range transport

North African dust reaches the southeastern United States every summer. Size-resolved dust mass measurements taken in Miami, Florida, indicate that more than one-half of the surface dust mass concentrations reside in particles with geometric diameters less than 2.1 µm, while vertical profiles of micropulse lidar depolarization ratios show dust reaching above 4 km during pronounced events. These observations are compared to the representation of dust in the Modern-Era Retrospective analysis for Research and Applications, version 2 (MERRA-2) aerosol reanalysis and closely related Goddard Earth Observing System model version 5 (GEOS-5) Forward Processing (FP) aerosol product, both of which assimilate satellite-derived aerosol optical depths using a similar protocol and inputs. These capture the day-to-day variability in aerosol optical depth well, in a comparison to an independent sun-photometer-derived aerosol optical depth dataset. Most of the modeled dust mass resides in diameters between 2 and 6 µm, in contrast to the measurements. Model-specified mass extinction efficiencies equate light extinction with approximately 3 times as much aerosol mass, in this size range, compared to the measured dust sizes. GEOS-5 FP surface-layer sea salt mass concentrations greatly exceed observed values, despite realistic winds and relative humidities. In combination, these observations help explain why, despite realistic total aerosol optical depths, (1) free-tropospheric model volume extinction coefficients are lower than those retrieved from the micro-pulse lidar, suggesting too-low model dust loadings in the free troposphere, and (2) model dust mass concentrations near the surface can be higher than those measured. The modeled vertical distribution of dust, when captured, is reasonable. Large, aspherical particles exceeding the modeled dust sizes are also occasionally present, but dust particles with diameters exceeding 10 µm contribute little to the measured total dust mass concentrations after such long-range transport. Remaining uncertainties warrant a further integrated assessment to confirm this study's interpretations.

54 ENVIRONMENTAL SCIENCES↗

Effects of fast and thermal neutron irradiation on Ga-polar and N-polar GaN diodes

Studies of the radiation tolerance and electrical behavior of gallium nitride (GaN) based devices are important for the next generation of high-power and high-voltage electronics that may be subjected to harsh environments such as nuclear reactor and fusion facilities, particle accelerators, and post-denotation environments. Here, in this work, we study the behavior of Ga-polar and N-polar GaN Schottky diodes before and after exposure to fast and thermal + fast neutrons. Temperature-dependent current–voltage (I–V) and circular transmission line method (CTLM) measurements were used to study the electrical characteristics. A strong reduction in reverse leakage current and an increase in differential resistance in forward bias were observed after neutron irradiation. Thermionic emission (TE), Frenkel–Poole (FP) emission, and Fowler–Nordheim (FN) tunneling models were used to explain the forward and reverse I–V characteristics pre- and post-irradiation. The study confirms that Ga-polar and N-polar GaN Schottky diodes exhibit different electrical responses to fast and thermal neutron irradiations. The reverse bias characteristics of N-polar diodes are less affected after the fast neutron irradiation compared to Ga-polar diodes, while in the forward bias region, the electrical behavior after fast and thermal neutron irradiations is similar in Ga-polar and N-polar diodes. The results indicate that the role of orientation should be considered in the design of GaN-based radiation-tolerant electronics.

42 ENGINEERING↗

PINNSTRIPES (Physics-Informed Neural Network SurrogaTe for Rapidly Identifying Parameters in Energy Systems) [SWR-22-12]

Energy systems models typically take the form of complex partial differential equations which make multiple forward calculations prohibitively expensive. Fast and data-efficient construction of surrogate models is of utmost importance for applications that require parameter exploration such as design optimization and Bayesian calibration. In presence of a large number of parameters, surrogate models that capture correct dependencies may be difficult to construct with traditional techniques. The issue is addressed here with the formulation of the surrogate model constructed via Physics-Informed Neural Networks (PINN) which capture the dependence with respect to the parameters to estimate, while using a limited amount of data. Since forward evaluations of the surrogate model are cheap, parameter exploration is made inexpensive, even when considering a large number of parameters.

Hassanaly, Malik↗

Uncertainty quantification in reacting flow.

Chemically reacting flow models generally involve inputs and parameters that are determined from empirical measurements, and therefore exhibit a certain degree of uncertainty. Estimating the propagation of this uncertainty into computational model output predictions is crucial for purposes of reacting flow model validation, model exploration, as well as design optimization. Recent years have seen great developments in probabilistic methods and tools for efficient uncertainty quantification (UQ) in computational models. These tools are grounded in the use of Polynomial Chaos (PC) expansions for representation of random variables. The utility and effectiveness of PC methods have been demonstrated in a range of physical models, including structural mechanics, transport in porous media, fluid dynamics, aeronautics, heat transfer, and chemically reacting flow. While high-dimensionality remains nominally an ongoing challenge, great strides have been made in dealing with moderate dimensionality along with non-linearity and oscillatory dynamics. In this talk, I will give an overview of UQ in chemical systems. I will cover both: (1) the estimation of uncertain input parameters from empirical data, and (2) the forward propagation of parametric uncertainty to model outputs. I will cover the basics of forward PC UQ methods with examples of their use. I will also highlight the need for accurate estimation of the joint probability density over the uncertain parameters, in order to arrive at meaningful estimates of model output uncertainties. Finally, I will discuss recent developments on the inference of this density given partial information from legacy experiments, in the absence of raw data.

99 GENERAL AND MISCELLANEOUS↗

Uncertainty quantification in reacting flow.

Chemically reacting flow models generally involve inputs and parameters that are determined from empirical measurements, and therefore exhibit a certain degree of uncertainty. Estimating the propagation of this uncertainty into computational model output predictions is crucial for purposes of reacting flow model validation, model exploration, as well as design optimization. Recent years have seen great developments in probabilistic methods and tools for efficient uncertainty quantification (UQ) in computational models. These tools are grounded in the use of Polynomial Chaos (PC) expansions for representation of random variables. The utility and effectiveness of PC methods have been demonstrated in a range of physical models, including structural mechanics, transport in porous media, fluid dynamics, aeronautics, heat transfer, and chemically reacting flow. While high-dimensionality remains nominally an ongoing challenge, great strides have been made in dealing with moderate dimensionality along with non-linearity and oscillatory dynamics. In this talk, I will give an overview of UQ in chemical systems. I will cover both: (1) the estimation of uncertain input parameters from empirical data, and (2) the forward propagation of parametric uncertainty to model outputs. I will cover the basics of forward PC UQ methods with examples of their use. I will also highlight the need for accurate estimation of the joint probability density over the uncertain parameters, in order to arrive at meaningful estimates of model output uncertainties. Finally, I will discuss recent developments on the inference of this density given partial information from legacy experiments, in the absence of raw data.

99 GENERAL AND MISCELLANEOUS↗

Energize New York -- Residential Energy Efficiency Market Transformation in New York

Energize New York (ENY) is focused on addressing known barriers to widespread energy efficiency (EE) adoption by property owners. In simple terms ENY works to; 1) increase homeowner trust by using known and trusted community leaders to communicate their own positive experiences with EE and ENY, 2) reduce friction points by integrating support staff and creative use of technology that ease a homeowner’s way forward through the EE upgrade process, 3) increase homeowner knowledge and trust of specific EE benefits through access to the ENY energy coach, 4) provide information on available financing options that show benefits (e.g. positive cash flow), and 5) provide tools and mechanisms that give homeowners greater comfort in evaluating and selecting the right contractor. These five fundamental program aspects are supported by a hyper-local communications and outreach model and a “lead by example” philosophy that requires community leaders to step forward and model energy efficiency behavior by example. In the communities where the program has successfully engaged local leadership and “trusted community sources” there has been a significant increase in the uptake of energy efficiency work. Quantitatively this growth translated into an increase of 240% in completed Home Performance projects from the two year pre-pilot period (2009-10) to the two year post-pilot period (2012-13). Additionally, the program has seen measureable increases in the output and performance of contractors who are members of the Energize Comfort Corps, a program innovation which included a subset of the NYSERDA approved and Building Performance Institute accredited home performance contractors. This innovation, launched in the later stages of the pilot, is paired with the Energize Contractor Ratings Index (CRI) and helps frame the provider market by giving homeowners a mechanism to provide and receive feedback on their contractor experience. The data incorporated into the CRI along with the ECC selection process, has measurably reduced a key decision barrier (“which contractor should I use”) for homeowners engaged in the EE decision making process.

32 ENERGY CONSERVATION, CONSUMPTION, AND UTILIZATI↗

Delineating Variabilities of Groundwater Level Prediction Across the Agriculturally Intensive Transboundary Aquifers of South Asia

We report the groundwater depletion in South Asia’s Himalayan, transboundary Indus-Ganges-Brahmaputra-Meghna (IGBM) rivers basin is among the highest globally. Given the high irrigation demand and population, groundwater sustainability requires an improved understanding of groundwater systems for the accurate prediction of groundwater levels (GWLs). However, the prediction of groundwater system behaviors is a significant challenge since it is dominated by spatiotemporal and subsurface depth-dependent drivers. Earlier studies that address the challenges are mainly based on the short spatial and temporal extent and/or do not separate the renewable (i.e., shallow) vs nonrenewable (i.e., deeper) groundwater signals. Here, we first identified the variable importance of spatial and depth-dependent drivers on GWL in the IGBM basin. Our results indicate a greater influence of anthropogenic factors (i.e., widespread pumping and increased population) in most parts of the IGBM basin, except in the precipitation-dominated basin of the Brahmaputra. Our next purpose was to delineate a multifactorial approach for GWL prediction using the two most used machine learning models (i.e., support vector machine and feed-forward neural network) in the literature. In general, the machine learning model outputs show a good match in comparison to the GWL from the observation wells (n = 2303 distributed across India and Bangladesh) with some limitations in areas with increased groundwater irrigation. We separately compared the results from shallow (<35 m) and deep (>35 m) observation wells, emphasizing the significance of deep groundwater pumping. Our approach highlights the importance of spatiotemporal to multidepth factors in GWL prediction and can be adopted in other parts of the globe to predict GWLs.

54 ENVIRONMENTAL SCIENCES↗

A forecast-driven decision-making model for long-term operation of a hydro-wind-photovoltaic hybrid system

Hydro-wind-photovoltaic (PV) hybrid system has the potential to increase the integration of renewable energy sources into an existing grid. For the long-term operation of the system, due to the non-storable nature of wind and PV power, it is essentially to decide the optimal long-term carryover storage of cascade reservoirs. However, it remains a challenge due to high uncertainties of long-term forecasts and complicated hydraulic/electrical relationships between cascade reservoirs. Here in this study, a forecast-driven decision-making model is proposed for the hybrid system, which converts the multi-stage long-term operation process into a two-stage operation problem (including current stage and carryover stage) to avoid using longer-horizon forecast information with lower accuracy. First, the carryover stage energy surfaces (CESs) considering the forecast uncertainties of wind, solar and hydro resources are proposed to characterize carryover stage benefit quantitatively. Then a CESs-based forward decision-making optimization model is developed to guide the long-term operation of a hydro-wind-photovoltaic hybrid system. The applications in a hydro-wind-PV hybrid system of Yalong River basin results show that: compared with conventional operation, 1) power generation increases 9.03%; 2) in terms of the carryover storages control, the reservoir impounding and drawdown timing are delayed, and the drawdown depth is increased, which can be used to formulate better reservoir operation rules.

13 HYDRO ENERGY↗

Reactor antineutrino flux and anomaly

Reactor antineutrinos have played a significant role in establishing the standard model of particle physics and the theory of neutrino oscillations. In this article, we review the reactor antineutrino flux and in particular the reactor antineutrino anomaly (RAA) coined over a decade ago. RAA refers to a deficit of the measured antineutrino inverse beta decay rates at very short-baseline reactor experiments compared to the theoretically improved predictions (i.e. the Huber–Mueller model). Since the resolution of several previous experimental anomalies have led to the discovery of non-zero neutrino mass and mixing, many efforts have been invested to study the origin of RAA both experimentally and theoretically. The progress includes the observation of discrepancies in antineutrino energy spectrum between data and the Huber–Mueller model, the re-evaluation of the Huber–Mueller model uncertainties, the potential isotope-dependent rate deficits, and the better agreement between data and new model predictions using the improved summation method. Importantly, these developments disfavor the hypothesis of a light sterile neutrino as the explanation of RAA and supports the deficiencies of Huber–Mueller model as the origin. Looking forward, more effort from both the theoretical and experimental sides is needed to fully understand the root of RAA and to make accurate predictions of reactor antineutrino flux and energy spectrum for future discoveries.

72 PHYSICS OF ELEMENTARY PARTICLES AND FIELDS↗

Forward problem solution as the operator of filtered and back projection matrix to reconstruct the various method of data collection and the object element model in electrical impedance tomography

Back projection reconstruction has been implemented to get the dynamical image in electrical impedance tomography. However the implementation is still limited in method of adjacent data collection and circular object element model. The study aims to develop the methods of back projection as reconstruction method that has the high speed, accuracy, and flexibility, which can be used for various methods of data collection and model of the object element. The proposed method uses the forward problem solution as the operator of filtered and back projection matrix. This is done through a simulation study on several methods of data collection and various models of the object element. The results indicate that the developed method is capable of producing images, fastly and accurately for reconstruction of the various methods of data collection and models of the object element.

71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSIC↗

Conditional Karhunen–Loève regression model with Basis Adaptation for high-dimensional problems: Uncertainty quantification and inverse modeling

Here, we propose a methodology for improving the accuracy of surrogate models of the observable response of physical systems as a function of the systems’ spatially heterogeneous parameter fields, with applications to uncertainty quantification and parameter estimation in high-dimensional problems. Practitioners often formulate finite-dimensional representations of spatially heterogeneous parameter fields using truncated unconditional Karhunen–Loève expansions (KLEs) for a certain choice of unconditional covariance kernel and construct surrogate models of the observable response with respect to the KLE coefficients. When direct measurements of the parameter fields are available, we propose improving the accuracy of these surrogate models by representing the parameter fields via conditional Karhunen-Loève expansions (CKLEs). CKLEs are constructed by conditioning the covariance kernel of the unconditional expansion on the direct measurements of the parameter field via Gaussian process regression, and then truncating the corresponding KLE. We apply the proposed methodology to constructing surrogate models via the Basis Adaptation (BA) method of the stationary hydraulic head response, measured at spatially discrete observation locations, of a groundwater flow model of the Hanford Site, as a function of the 1000-dimensional representation of the model’s log-transmissivity field. We find that BA surrogate models of the hydraulic head based on CKLEs are more accurate than BA surrogate models based on unconditional expansions for forward uncertainty quantification tasks. Furthermore, we find that inverse estimates of the hydraulic transmissivity field computed using CKLE-based BA surrogate models are more accurate than those computed using unconditional BA surrogate models.

97 MATHEMATICS AND COMPUTING↗

Fracture Network and Fluid Flow Imaging for Enhanced Geothermal Systems Applications from Multi-Dimensional Electrical Resistivity Structure

We have developed an algorithm for the inversion of magnetotelluric (MT) data to a 3D earth resistivity model based upon the finite element method. Hexahedral edge finite elements are implemented to accommodate discontinuities in the electric field across resistivity boundaries, and to accurately simulate topographic variations. All matrices are reduced and solved using direct solution modules which avoids ill-conditioning endemic to iterative solvers such as conjugate gradients, principally PARDISO for the finite element system and PLASMA for the parameter step estimate. Large model parameterizations can be handled by transforming the Gauss-Newton estimator to data-space form. Accuracy of the forward problem and jacobians has been checked by comparison to integral equations results and by limiting asymptotes. Inverse accuracy and performance has been verified against the public Dublin Secret Test Model 2 and the well-known Mount St Helens 3D MT data set. This algorithm we believe is the most capable yet for forming 3D images of earth resistivity structure and their implications for geothermal fluids and pathways.

15 GEOTHERMAL ENERGY↗

Measurement and modeling of intrinsic transcription terminators

The reliable forward engineering of genetic systems remains limited by the ad hoc reuse of many types of basic genetic elements. Although a few intrinsic prokaryotic transcription terminators are used routinely, termination efficiencies have not been studied systematically. Here, we developed and validated a genetic architecture that enables reliable measurement of termination efficiencies. We then assembled a collection of 61 natural and synthetic terminators that collectively encode termination efficiencies across an 800-fold dynamic range within Escherichia coli. We simulated co-transcriptional RNA folding dynamics to identify competing secondary structures that might interfere with terminator folding kinetics or impact termination activity. We found that structures extending beyond the core terminator stem are likely to increase terminator activity. By excluding terminators encoding such context-confounding elements, we were able to develop a linear sequence-function model that can be used to estimate termination efficiencies (r = 0.9, n = 31) better than models trained on all terminators (r = 0.67, n = 54). The resulting systematically measured collection of terminators should improve the engineering of synthetic genetic systems and also advance quantitative modeling of transcription termination.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗

Bayesian uncertainty quantification of tristructural isotropic particle fuel silver release: Decomposing model inadequacy plus experimental noise and parametric uncertainties

Tristructural isotropic (TRISO) particle fuel is one of the most promising fuel concepts enabling high temperature and high burnup reactor operation. One dominant source of radioactivity released from the TRISO particles is silver (Ag), which is subject to a high release fraction and long decay life compared to other fission products. Previous modeling efforts using the fuel performance code BISON indicated nonnegligible uncertainties in modeling the diffusion process of fission products in TRISO compared to the Advanced Gas Reactor experiments. The overall uncertainties observed when modeling the fission product diffusion can result from uncertainties in model parameters, noisy experimental measurements, and deficiencies in the developed models. The three types of underlying uncertainties have not yet been properly quantified in open literature. Here, this paper presents the Bayesian uncertainty quantification (UQ) using massively parallelizable Markov chain Monte Carlo samplers. The uncertainties due to model parameters, model inadequacy, and experimental measurement noise are quantified, with the σ term used to represent the sum of the model inadequacy and measurement noise uncertainties. It is worth noting that this is the first time the σ term is inferred for nuclear fuel experiments, as compared to using prescribed values for uncertainty quantification in previous work. The parallelizable Markov chain Monte Carlo samplers efficiently infer the model parameters and the σ term, giving insight into physical parameters like diffusion coefficients and the combined model discrepancy and measurement noise. A subsequent forward uncertainty quantification (UQ) is also performed based on the calibration results to generate more accurate predictions of the Ag release. The model inadequacy plus experimental noise is the most dominant source of uncertainty compared to the parametric uncertainty. All the UQ analyses presented in this work are based on the second series of the irradiation experiments in the Advanced Gas Reactor program.

11 NUCLEAR FUEL CYCLE AND FUEL MATERIALS↗

The afterglow of GRB 130427A from 1 to 10{sup 16} GHz

We present multiwavelength observations of the afterglow of GRB 130427A, the brightest (in total fluence) gamma-ray burst (GRB) of the past 29 yr. Optical spectroscopy from Gemini-North reveals the redshift of the GRB to be z = 0.340, indicating that its unprecedented brightness is primarily the result of its relatively close proximity to Earth; the intrinsic luminosities of both the GRB and its afterglow are not extreme in comparison to other bright GRBs. We present a large suite of multiwavelength observations spanning from 300 s to 130 days after the burst and demonstrate that the afterglow shows relatively simple, smooth evolution at all frequencies, with no significant late-time flaring or rebrightening activity. The entire data set from 1 GHz to 10 GeV can be modeled as synchrotron emission from a combination of reverse and forward shocks in good agreement with the standard afterglow model, providing strong support to the applicability of the underlying theory and clarifying the nature of the GeV emission observed to last for minutes to hours following other very bright GRBs. A tenuous, wind-stratified circumburst density profile is required by the observations, suggesting a massive-star progenitor with a low mass-loss rate, perhaps due to low metallicity. GRBs similar in nature to GRB 130427A, inhabiting low-density media and exhibiting strong reverse shocks, are probably not uncommon but may have been difficult to recognize in the past owing to their relatively faint late-time radio emission; more such events should be found in abundance by the new generation of sensitive radio and millimeter instruments.

79 ASTRONOMY AND ASTROPHYSICS↗

Self-Supervised Cloud Classification

Abstract Low-level marine clouds play a pivotal role in Earth’s weather and climate through their interactions with radiation, heat and moisture transport, and the hydrological cycle. These interactions depend on a range of dynamical and microphysical processes that result in a broad diversity of cloud types and spatial structures, and a comprehensive understanding of cloud morphology is critical for continued improvement of our atmospheric modeling and prediction capabilities moving forward. Deep learning has recently accelerated our ability to study clouds using satellite remote sensing, and machine learning classifiers have enabled detailed studies of cloud morphology. A major limitation of deep learning approaches to this problem, however, is the large number of hand-labeled samples that are required for training. This work applies a recently developed self-supervised learning scheme to train a deep convolutional neural network (CNN) to map marine cloud imagery to vector embeddings that capture information about mesoscale cloud morphology and can be used for satellite image classification. The model is evaluated against existing cloud classification datasets and several use cases are demonstrated, including training cloud classifiers with very few labeled samples, interrogation of the CNN’s learned internal feature representations, cross-instrument application, and resilience against sensor calibration drift and changing scene brightness. The self-supervised approach learns meaningful internal representations of cloud structures and achieves comparable classification accuracy to supervised deep learning methods without the expense of creating large hand-annotated training datasets. Significance Statement Marine clouds heavily influence Earth’s weather and climate, and improved understanding of marine clouds is required to improve our atmospheric modeling capabilities and physical understanding of the atmosphere. Recently, deep learning has emerged as a powerful research tool that can be used to identify and study specific marine cloud types in the vast number of images collected by Earth-observing satellites. While powerful, these approaches require hand-labeling of training data, which is prohibitively time intensive. This study evaluates a recently developed self-supervised deep learning method that does not require human-labeled training data for processing images of clouds. We show that the trained algorithm performs competitively with algorithms trained on hand-labeled data for image classification tasks. We also discuss potential downstream uses and demonstrate some exciting features of the approach including application to multiple satellite instruments, resilience against changing image brightness, and its learned internal representations of cloud types. The self-supervised technique removes one of the major hurdles for applying deep learning to very large atmospheric datasets.

54 ENVIRONMENTAL SCIENCES↗