Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “Multivariate regression”

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 235 records · Page 13
All sourcesNASA NTRSDOE OSTI
DOE OSTI2021

Identification and correction of temporal and spatial distortions in scanning transmission electron microscopy

Scanning transmission electron microscopy (STEM) has become the technique of choice for quantitative characterization of atomic structure of materials, where the minute displacements of atomic columns from high-symmetry positions can be used to map strain, polarization, octahedra tilts, and other physical and chemical order parameter fields. The latter can be used as inputs into mesoscopic and atomistic models, providing insight into the correlative relationships and generative physics of materials on the atomic level. However, these quantitative applications of STEM necessitate understanding the microscope induced image distortions and developing the pathways to compensate them both as part of a rapid calibration procedure for in situ imaging, and the post-experimental data analysis stage. Here, we explore the spatiotemporal structure of the microscopic distortions in STEM using multivariate analysis of the atomic trajectories in the image stacks. Based on the behavior of principal component analysis (PCA), we develop the Gaussian process (GP)-based regression method for quantification of the distortion function. The limitations of such an approach and possible strategies for implementation as a part of in-line data acquisition in STEM are discussed. Here, the analysis workflow is summarized in a Jupyter notebook that can be used to retrace the analysis and analyze the reader's data.

36 MATERIALS SCIENCE↗
DOE OSTI2022

Comparing Calibration Algorithms for the Rapid Characterization of Pretreated Corn Stover Using Near-Infrared Spectroscopy

Rapid characterization of biomass composition is a key enabling technology for biorefineries—the ability to measure the chemical composition of biomass materials entering the biorefinery as well as the composition of key process intermediate streams would allow real-time process control and the development of robust models to predict process performance. The utility of near-infrared (NIR) spectroscopy for rapid characterization requires multivariate algorithms for building calibration models. The most prevalent algorithm used for building calibration models using NIR spectra is the linear modeling algorithm Partial Least Squares Regression (PLS). Nonlinear regression algorithms (which are typically more computationally intensive than linear modeling approaches) have gained popularity in recent years due to their ability to solve a wide variety of classification and regression problems and the dramatic increase in available computational resources. In this work, we demonstrate that a calibration model can predict the composition of corn stover process intermediate samples pretreated with three different treatments—hot water (HW), dilute acid (DA), and deacetylation followed by dilute acid (DDA). We quantitatively compare three different algorithms for building prediction models based on near-infrared spectroscopy—partial least squares (PLS), support vector machines (SVM), and random forests (RF). We demonstrate the utility of improving model performance by accounting for instrument performance variability using repeated measurements of standard materials (e.g., the “repeatability file” strategy) and investigate its performance with nonlinear regression techniques, and we discuss methods for quantifying the uncertainties of specific predictions among the three methods.

09 BIOMASS FUELS↗
DOE OSTI2023

Hierarchical Modeling to Enhance Spectrophotometry Measurements—Overcoming Dynamic Range Limitations for Remote Monitoring of Neptunium

A robust hierarchical model has been demonstrated for monitoring a wide range of neptunium concentrations (0.75–890 mM) and varying temperatures (10–80 °C) using chemometrics and feature selection. The visible–near infrared electronic absorption spectrum (400–1700 nm) of monocharged neptunyl dioxocation (Np(V) = NpO2+) includes many bands, which have molar absorption coefficients that differ by nearly 2 orders of magnitude. The shape, position, and intensity of these bands differ with chemical interactions and changing temperature. These challenges make traditional quantification by univariate methods unfeasible. Measuring Np(V) concentration over several orders of magnitude would typically necessitate cells with varying path length, optical switches, and/or multiple spectrophotometers. Alternatively, the differences in the molar extinction coefficients for multiple absorption bands can be used to quantify Np(V) concentration over 3 orders of magnitude with a single optical path length (1 mm) and a hierarchical multivariate model. In this work, principal component analysis was used to distinguish the concentration regime of the sample, directing it to the relevant partial least squares regression submodels. Each submodel was optimized with unique feature selection filters that were selected by a genetic algorithm to enhance predictions. Through this approach, the percent root mean square error of prediction values were ≤1.05% for Np(V) concentrations and ≤4% for temperatures. This approach may be applied to other nuclear fuel cycle and environmental applications requiring real-time spectroscopic measurements over a wide range of conditions.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗
DOE OSTI2020

Combining Emergent Constraints for Climate Sensitivity

A method is proposed for combining information from several emergent constraints into a probabilistic estimate for a climate sensitivity proxy Y such as equilibrium climate sensitivity (ECS). The method is based on fitting a multivariate Gaussian PDF for Y and the emergent constraints using an ensemble of global climate models (GCMs); it can be viewed as a form of multiple linear regression of Y on the constraints. The method accounts for uncertainties in sampling this multidimensional PDF with a small number of models, for observational uncertainties in the constraints, and for overconfidence about the correlation of the constraints with the climate sensitivity. Its general form (Method C) accounts for correlations between the constraints. Method C becomes less robust when some constraints are too strongly related to each other; this can be mitigated using regularization approaches such as ridge regression. An illuminating special case, Method U, neglects any correlations between constraints except through their mutual relationship to the climate proxy; it is more robust to small GCM sample size and is appealingly interpretable. These methods are applied to ECS and the climate feedback parameter using a previously published set of 11 possible emergent constraints derived from climate models in the Coupled Model Intercomparison Project (CMIP). The ±2σ posterior range of ECS for Method C with no overconfidence adjustment is 4.3 ± 0.7 K. For Method U with a large overconfidence adjustment, it is 4.0 ± 1.3 K. This study adds confidence to past findings that most constraints predict higher climate sensitivity than the CMIP mean.

54 ENVIRONMENTAL SCIENCES↗
DOE OSTI2023

Online evolutionary neural architecture search for multivariate non-stationary time series forecasting

Time series forecasting (TSF) is one of the most important tasks in data science. TSF models are usually pre-trained with historical data and then applied on future unseen datapoints. However, real-world time series data is usually non-stationary and models trained offline usually face problems from data drift. Models trained and designed in an offline fashion can not quickly adapt to changes quickly or be deployed in real-time. To address these issues, this work presents the Online NeuroEvolution-based Neural Architecture Search (ONE-NAS) algorithm, which is a novel neural architecture search method capable of automatically designing and dynamically training recurrent neural networks (RNNs) for online forecasting tasks. Without any pre-training, ONE-NAS utilizes populations of RNNs that are continuously updated with new network structures and weights in response to new multivariate input data. ONE-NAS is tested on real-world, large-scale multivariate wind turbine data as well as the univariate Dow Jones Industrial Average (DJIA) dataset. These results demonstrate that ONE-NAS outperforms traditional statistical time series forecasting methods, including online linear regression, fixed long short-term memory (LSTM) and gated recurrent unit (GRU) models trained online, as well as state-of-the-art, online ARIMA strategies. Additionally, results show that utilizing multiple populations of RNNs which are periodically repopulated provide significant performance improvements, allowing this online neural network architecture design and training to be successful.

97 MATHEMATICS AND COMPUTING↗
NASA NTRS1989

Remote sensing of geobotanical trends in East Africa

The spatial and phenological distribution of vegetation was examined in a remote sensing geobotanical study of the East Africa Rift region. Six normalized difference vegetation index (NDVI) AVHRR (advanced very high resolution radiometer) scenes (February, June, and September of both 1984 and 1987) were used as a measure of vegetation presence. NOAA ETOPO-5 elevation and geographic coordinate data were coregistered with the multitemporal NDVI images. Univariate and multivariate statistics indicate that the NDVI values are significantly associated with elevation, latitude, and longitude. This supports the concept of quantifiable, regional gradients that affect large-scale geobotanical studies. A quadratic regression line was fitted to the NDVI, elevation, latitude, and longitudinal data. In this way a regional trend, as affected by elevation, was determined. The deviation of the actual data from this regional trend was displayed as an image, and shows the local climatic and geological influences.

Warner, Timothy A.↗
DOE OSTI2022

Effect of vaccination on the case fatality rate for COVID-19 infections 2020–2021: multivariate modelling of data from the US Department of Veterans Affairs

Objectives: To evaluate the benefits of vaccination on the case fatality rate (CFR) for COVID-19 infections. Design, setting and participants: The US Department of Veterans Affairs has 130 medical centres. We created multivariate models from these data—339 772 patients with COVID-19—as of 30 September 2021. Outcome measures: The primary outcome for all models was death within 60 days of the diagnosis. Logistic regression was used to derive adjusted ORs for vaccination and infection with Delta versus earlier variants. Models were adjusted for confounding factors, including demographics, comorbidity indices and novel parameters representing prior diagnoses, vital signs/baseline laboratory tests and outpatient treatments. Patients with a Delta infection were divided into eight cohorts based on the time from vaccination to diagnosis. A common model was used to estimate the odds of death associated with vaccination for each cohort relative to that of unvaccinated patients. Results: 9.1% of subjects were vaccinated. 21.5% had the Delta variant. 18 120 patients (5.33%) died within 60 days of their diagnoses. The adjusted OR for a Delta infection was 1.87±0.05, which corresponds to a relative risk (RR) of 1.78. The overall adjusted OR for prior vaccination was 0.280±0.011 corresponding to an RR of 0.291. Raw CFR rose steadily after 10–14 weeks. The OR for vaccination remained stable for 10–34 weeks. Conclusions: Our CFR model controls for the severity of confounding factors and priority of vaccination, rather than solely using the presence of comorbidities. Our results confirm that Delta was more lethal than earlier variants and that vaccination is an effective means of preventing death. After adjusting for major selection biases, we found no evidence that the benefits of vaccination on CFR declined over 34 weeks. We suggest that this model can be used to evaluate vaccines designed for emerging variants.

59 BASIC BIOLOGICAL SCIENCES↗
NASA NTRS2021

Probabilistic Machine Learning Estimation of Ocean Mixed Layer Depth from Dense Satellite and Sparse In-Situ Observations

The ocean mixed layer plays an important role in the coupling between the upper ocean and atmosphere across a wide range of time scales. Estimation of the variability of the ocean mixed layer is therefore important for atmosphere-ocean prediction and analysis. The increasing coverage of in situ Argo profile data allows for an increasingly accurate analysis of the mixed layer depth (MLD) variability associated with deviations from the seasonal climatology. However, sampling rates are not sufficient to fully resolve subseasonal (<90 day) MLD variability. Yet, many multivariate observations-based analyses include implicit modeled subseasonal MLD variability. One analysis method is optimal interpolation of in situ data, but the interior analysis can be improved by leveraging surface data with regression or variational approaches. Here, we demonstrate how machine learning methods and satellite sea surface temperature, salinity, and height facilitate MLD estimation in a pilot study of two regions: the mid-latitude southern Indian and the eastern equatorial Pacific Oceans. We construct multiple machine learning architectures to produce weekly 1/2° gridded MLD anomaly fields (relative to a monthly climatology) with uncertainty estimates. We test multiple traditional and probabilistic machine learning techniques to compare both accuracy and probabilistic calibration. We validate our methodology by applying it to ocean model simulations. We find that incorporating sea surface data through a machine learning model improves the performance of spatiotemporal MLD variability estimation compared to optimal interpolation of Argo observations alone. These preliminary results are a promising first step for the application of machine learning to MLD prediction.

Machine Learning↗
NASA NTRS2022

Bayesian Rules of Thumb: Robust Uncertainty Quantification in Early Project Cost Estimation

Systems engineers often make use of cost Rules ofThumb in order to estimate cost during early phases of projectformulation. These Rules of Thumb typically take the form ofa sequence of percentages over which a total cost is allocatedacross NASA WBS elements. Rules of Thumb can then be usedto extrapolate cost from one or more known WBS elements tothe remaining unknown WBS elements, assisting early projectformulation architecture studies (such as those in JPL’s Team Xand A Team).A number of issues can arise when generating and using costRules of Thumb. For example, many records of project costsconsist of incomplete data. Typical methods of dealing withincomplete cost allocation data include (a) ignoring missionswith incomplete data, or (b) taking averages of the non-zero percentagesacross missions, but both of these methods can result inbiased estimates if the existence of incomplete data correlateswith total mission cost or any particular WBS element. Anothercommon example is cost reported in one or more incorrect WBSelements. This is especially prevalent in smaller missions whereit is more common for engineers to perform tasks that fall underthe purview of multiple WBS elements.Furthermore, a Rule of Thumb estimate is typically reported asa point estimate; there is no reported uncertainty around thepercentages used to generate an allocation. Even in the rarecase in which confidence intervals around mean percentages areprovided, there may be positive or negative correlations betweenWBS elements which can skew estimates.Here we attempt to address these problems by formulatingprobabilistic Rules of Thumb in which a distribution of allocationschemes, rather than a single allocation scheme, is generated.We use a bootstrap imputation method to simultaneouslyaccount for uncertainty in the missing data while using allavailable information contained in the dataset. The imputeddatasets are then input into a multivariate Bayesian modelwhich accounts for correlations between WBS elements andproperly accounts for uncertainty in the final Rule of Thumbpercentages and predictions. We describe the mathematicalmodel and provides snippets of R code utilizing the brms(Bayesian Regression Models using Stan) package. To illustratethis model, we generate a Bayesian Level 2 WBS Cost Rule ofThumb for MIDEX (Medium-Class Explorers) missions withdata extracted from NASA’s CADRe. We then compare thismethod’s performance with the classical Rule of Thumb method.

Hooke, Melissa A↗
NASA NTRS1993

Application of a two-stream radiative transfer model for leaf lignin and cellulose concentrations from spectral reflectance measurements, part 1

Lignin and nitrogen contents of leaves constitute the primary rate-limiting parameters for the decomposition of forest litter, and are determinants of nutrient- and carbon-cyclic rates in forest ecosystems (Melillo et al., 1982). Wessman et al. (1988a) developed empirical multivariate relationships between forest canopy lignin amount and the (first-difference) AIS spectral response in three bands spread over the wavelength interval 1256-1555 nm. Wessman et al. (1988b) and McLellan et al. (1991) developed similar regression relationships from laboratory reflectance measurements on dried samples prepared in a standard fashion. They used four to six infrared bands for analysis of nitrogen, lignin and cellulose content of foliage in forest and prairie species. In the present article (Parts 1 and 2) the feasibility of compositional determinations is explored using positions of composite absorption bands that originate from mixtures of lignin, cellulose, and possibly other chemical constituents in the spectral reflectance of green leaves. To carry out this program, we employ full-spectral-resolution single-leaf diffuse reflectance measurements made with a laboratory spectrometer and integrating sphere. The leaf and other chemical reflectance data compiled by Elvidge (1990) have also been utilized extensively.

Conel, James E.↗
DOE OSTI2025

Joint Modeling of Quasar Variability and Accretion Disk Reprocessing Using Latent Stochastic Differential Equations

Quasars are bright active galactic nuclei powered by the accretion of matter around supermassive black holes at the center of galaxies. Their stochastic brightness variability depends on the physical properties of the accretion disk and black hole. The upcoming Rubin Observatory Legacy Survey of Space and Time (LSST) is expected to observe tens of millions of quasars, so there is a need for efficient techniques like machine learning that can handle the large volume of data. Quasar variability is believed to be driven by an X-ray corona, which is reprocessed by the accretion disk and emitted as UV/optical variability. We are the first to introduce an auto-differentiable simulation of the accretion disk and reprocessing. We use the simulation as a direct component of our neural network to jointly model the driving variability and reprocessing, trained with supervised learning on simulated LSST-like 10 yr quasar light curves. We encode the light curves using a transformer encoder, and the driving variability is reconstructed using latent stochastic differential equations, a physically motivated generative deep learning method that can model continuous-time stochastic dynamics. By embedding the physical processes of the driving signal and reprocessing into our network, we achieve a model that is more robust and interpretable. We demonstrate that our model outperforms a Gaussian process regression baseline and can infer accretion disk parameters and time delays between wave bands, even for out-of-distribution driving signals. Our approach provides a powerful framework that can be adapted to solve other inverse problems in multivariate time series.

Fagin, Joshua [City Univ. of New York (CUNY), NY (↗
NASA NTRS2022

Gaussian Process Regression Method for Costing SmallSat Bus Capabilities

NASA is responding to the growing interest in, andcapabilities of, small satellites for science applications with an increasingnumber and frequency of Announcements of Opportunityfor small satellite space missions. Estimating the probabilitythat these mission concepts will fit within the small cost capsof these opportunities is largely driven by the probability thatone of the burgeoning number of small satellite providers will beable to meet the payload’s accommodation requirements withinthe budget for the spacecraft. JPL has collected a databasecontaining technical specifications and cost of commerciallyavailable Smallsat buses across various vendors. The primarypurpose of the database is for use in JPL’s Team X architecturestudies to inform cost estimates of a spacecraft bus which fitsthe customer’s technical requirements for their payload andmission. Customer needs are often unique and don’t alignperfectly with an off-the-shelf commercial spacecraft bus, whichmotivates the need to develop a cost model across the continuoustechnical parameter space.Al’s Bus Cost Distribution Estimator (ABCDE) uses Gaussianprocess regression (GPR) to predict commercial Smallsat spacecraftbus cost based on a subset of a customer’s technicalrequirements (payload mass, payload power, delta V, pointingcontrol, and downlink rate). GPR is implemented in ABCDE asa Bayesian method which fits an implied multivariate regressionon the technical parameters and uses kriging to intentionally“overfit” the residuals. Overfitting the residuals allows costestimates to collapse in uncertainty closer to the data pointswhile maintaining larger uncertainty intervals in regions of parameterspace with fewer data records. The data used to fit thismodel is sensitive and represents cost estimates for off-the-shelfcommercial buses. GPR simultaneously protects the sensitivityof the database and uses the sparse nature of the database toaccount for uncertainty in cost in a useful way. For a givenset of customer technical requirements, the tool provides a costestimate distribution, the percentiles of which can be interpretedas a confidence level of finding a commercial bus under a specifiedcost cap. ABCDE dramatically pushes the boundaries ofspacecraft cost estimation models due to its Bayesian methodology(accounting for the maximum uncertainty in the underlyingregression), the mathematically advanced kriging methodology,and the novelty of its application in Team X architecture tradestudies.

Austin, Alex↗
DOE OSTI2024

Insights into Tetravalent Np Speciation in HNO 3 through Spectroelectrochemistry and Multivariate Analysis

In situ optical spectroscopy, spectropotentiometry, and multivariate analysis were applied to the Np(IV) nitrate system to better understand speciation and quantify HNO 3 concentration. Thin-layer spectropotentiometry, or spectroelectrochemistry, was leveraged to isolate and stabilize Np(IV) without compromising the solution conditions and generate representative Vis-NIR absorption spectra from 0.5 to 10 M HNO 3 and benchmark the corresponding Np(IV) molar absorptivity coefficients. Spectra were described with principal component analysis (PCA) to identify the purest Np(IV) absorbance spectra among other oxidation states [e.g., Np(V/VI)] at each acid concentration and then to identify the primary sources of variance within each Np(IV) spectrum with respect to Np(IV) nitrate complexes. Then, partial least-squares regression (PLSR) and support vector regression (SVR) models were built to predict HNO 3 concentration from the Np(IV) spectral data. The nonlinear SVR model outperformed the linear PLSR model for the HNO 3 concentration predictions. Finally, the inclusion of spectra collected in edge and center point HNO 3 concentrations in the calibration set was determined to be crucial for producing models with strong predictive capabilities. The multivariate approach used in this study makes it possible to quantify HNO 3 concentration solely based on Np(IV) absorption spectra, which is essential to quantifying processing streams in various online monitoring applications.

38 RADIATION CHEMISTRY, RADIOCHEMISTRY, AND NUCLEA↗
NASA NTRSDate not supplied

Nearest-Neighbor Machine Learning Feature Selection for Interpretation of Microbial Molecular Signatures from Isotope Ratio Mass Spectrometry Data

Mass spectrometry (MS) promises to be a powerful tool for potential biosignature detection during astrobiological missions on ocean worlds in our solar system. Accurate and generalizable machine learning methods could enhance science return on investment by predicting seawater chemistry and classifying isotopic biosignatures, either as a signature consistent with microbial life (biotic) or as a novelty (unclassified/unique). However, machine learning models are likely to be complex and involve interactions between MS features, making biosignatures difficult to interpret. Feature selection methods provide biological and chemical context that help interpret the mechanisms of machine learning models, but these methods also need the ability to detect complex interactions. Previously, we developed a machine learning feature selection algorithm called nearest-neighbor projected distance regression (NPDR) that has the ability to identify important model features that involve complex interactions and automatically reduce correlation and the dimensionality in a high-dimensional variable space. The standard distance metrics used in NPDR – Manhattan and Euclidean – assume the multivariate data are isotropic, which is often violated in real data due to differences in the covariance between variables. Thus, we extend NPDR to include a random forest distance, and other anisotropic distance metrics, for computing nearest neighbors. We also augment the isotope-ratio MS data with time-series features from the raw MS signal to improve biotic classification. We test NPDR on our novel experimental ocean world seawater analog MS data. We measure isotope fractionations of volatile CO 2 that could be measured in exospheres or plumes. Samples include baseline abiotic conditions using a range of possible seawater chemistry consistent with Europa and Enceladus, and biotic samples that include microbes in these seawaters. We use penalized NPDR with random forest proximity to identify interpretable microbial molecular signatures. We compare features with random forest importance, and we train a classifier that discriminates between biotic and abiotic samples with high accuracy. These ML-trained ocean-world analog MS data could be used to assist in identifying biosignatures during future missions.

geochemistry↗
DOE OSTI2023

Quantifying neptunium oxidation states in nitric acid through spectroelectrochemistry and chemometrics

Controlled-potential in situ thin-layer spectropotentiometry was leveraged to generate visible/near-infrared (VIS/NIR) absorption spectral data sets for the development of chemometric models to quantify Np(III/IV/V/VI) oxidation states in HNO 3 . This technology would be valuable in laboratory studies and when monitoring process solutions to guide feed adjustments for radiochemical separations—the performance of which depends on oxidation state. This approach successfully isolated and stabilized Np species in pure (~99%) oxidation states without compromising solution optical properties. Multivariate curve resolution–alternating least squares models were evaluated to resolve spectral and component concentrations from a scan that sequentially produced Np(VI), Np(V), Np(IV), and Np(III) spectra with mixtures of two valences at a time. Although it provided a useful approximation, the method was not able to quantitively resolve each component likely because of rotational ambiguity. Additionally, partial least squares regression models were built from artificial and electrochemically generated VIS/NIR spectral training sets to study the effect of interionic interactions on spectral characteristics. Models built with true Bi-chemical mixtures of coexisting Np oxidation states and spectra generated from additive combinations of pure end points had similar prediction performance. This methodology can be used to directly quantify Np concentration and the ratio of Np oxidation states and other actinides in remote settings such as hot cells.

38 RADIATION CHEMISTRY, RADIOCHEMISTRY, AND NUCLEA↗
DOE OSTI2025

Mapping wall-to-wall fractional cover of Arctic tundra plant functional types in Alaska using 20-m spatial resolution satellite imagery and harmonized plot observations

Estimates of fractional cover (fCover) across given land surfaces are used to assess, and often model, vegetation composition and diversity, which are crucial for understanding the health and functioning of terrestrial ecosystems. Remote sensing provides a useful means for scaling local, plot-measured fCover estimates to regional scales. Leveraging a recently synthesized and harmonized plot database, this study generated wall-to-wall maps of fCover for six Alaskan-Arctic plant functional types (PFT), including non-vascular plants, forbs, graminoids, and deciduous and evergreen shrubs, using 20-m satellite data (Sentinel-1, Sentinel-2, ArcticDEM) using a machine learning regression approach, specifically the random forest (RF) algorithm, which is well-suited for handling nonlinear relationships and high-dimensional satellite datasets. This study additionally addressed the spatio-temporal inconsistencies e.g., sampling scale, plot size, and collection year in plot measured fCover by adopting a multivariate outlier detection approach—Cook’s distance—to identify high-quality plots for model training and validation. Our approach achieves high accuracy (R 2 = 0.59–0.93, root mean squared errors = 0.02–0.10 for all PFTs) between plot-observed and satellite-derived fCover when using high-quality plot samples. The mapped fCover characterizes the spatial patterns of different PFTs across the tundra biome at a 20-m resolution, providing key information needed for improved representation of Arctic tundra vegetation in terrestrial biosphere models to better understand climate-vegetation feedback across the Arctic tundra.

Arctic tundra↗
NASA NTRSDate not supplied

A Novel Machine Learning Method for Surface PM2.5 Estimations from Geostationary Satellites

Particulate matter (PM) with a diameter of less or equal to 2.5 μm, known as PM , affects human health as it penetrates the respiratory system. The Environmental Protection Agency (EPA) measures the atmospheric concentration of PM using air quality monitors stationed throughout the Continental United States (CONUS). Such measurements are points on a spatial domain and therefore, might not be representative of the air quality at nearby areas considering that the composition of the atmosphere is highly variable from place to place. Satellite based AOD permits a spatially uniform means of estimating PM and new geostationary satellites provide high temporal and spatial resolution estimation of AOD. However, the concentration of PM is non-linearly dependent on other atmospheric parameters that include relative humidity, temperature, and height of the planetary boundary layer. This information may be estimated at similar spatial and temporal resolutions as AOD from numerical modeling such as from the National Oceanic and Atmospheric Administration’s (NOAA) High Resolution Rapid Refresh (HRRR) model which resolves near real-time atmospheric conditions over the CONUS. The estimation of PM concentration is a multi-parametric problem that considers the effect of temporal dependencies among the different parameters. Deep learning approaches are appropriate for such complex estimation problems as they intrinsically capture relations among multiple non-linear parameters. This study compares deep-learning methods to traditional regression analysis to demonstrate the capabilities of these methods in predicting PM2.5 concentrations. Additionally, a novel ensemble learning approach is employed to identify scientific processes that could further improve the estimation of PM concentration. Utilizing Long Short-Term Memory (LSTM) neural networks, which are suitable for multivariate time series estimation problems as they are capable of learning long-term dependencies, individual models are created for each EPA station and trained on the aforementioned dataset collocated over each station. Individual station models are merged if the model's performance is improved by reducing the root mean squared error (RMSE) metric. This ensemble training method ultimately reduces the RMSE value. Evaluation of these results provide insights into physical processes and related observable parameters that may contribute to PM concentrations. Identified parameters evaluated to be statistically different between the merged and unmerged models are expected to improve overall performance. These new parameters are then utilized for reevaluation of the deep learning methods with an extreme gradient boosting model with an RMSE of 5.5 providing the best results.

George Priftis↗