Search NASA⌕ Search

SEARCH · Search NASA

Results for “Linear 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 289 records · Page 16

Performance Comparison of Machine Learning Models for Ultrasonic Nondestructive Evaluation of Alkali-Silica Reaction in Concrete

Alkali-silica reaction (ASR) causes concrete degradation, leading to cracking, rebar corrosion, and reduced structural integrity, which raises safety concerns. Ultrasonic nondestructive evaluation (NDE) effectively assesses concrete properties and monitors ASR progression. However, its deployment and analysis require specialized expertise and subjective interpretation. As computational power increases, artificial intelligence (AI) and machine learning (ML) algorithms are increasingly being used to automate NDE data analysis across various industries for AI-assisted automation. Regulatory agencies are adapting to this technological shift, prompting a need to evaluate current ML technologies’ capabilities and limitations in assessing concrete material properties and damage. This report presents a comparative analysis of four ML regression models for predicting concrete material damage induced by ASR expansion using long-term ultrasonic data monitoring. The models investigated include linear regression (LR), support vector regression (SVR), shallow neural networks (NN), and deep neural networks (DNN). LR, SVR, and shallow NN models use features extracted from ultrasonic signals, whereas the DNN model processes time-domain ultrasonic signals and frequency spectra directly. The study systematically compared the models’ performance from various perspectives, including model input, prediction performance, and generalization ability. The findings indicate significant variability in model performance, with some ML algorithms achieving very high or very low prediction accuracy depending on the preprocessing and feature engineering (extraction and selection) applied. Key insights include the observation that shallow ML models (LR, SVR, and shallow NNs) require meticulous preprocessing and feature extraction to achieve high accuracy. In contrast, the DNN model, although it bypasses the need for feature engineering, necessitates extensive preprocessing to mitigate noise and computational demands. The SVR model emerged as the top performer among the shallow models, and the DNN model exhibited superior performance on specific datasets but struggled with generalization across specimens from different batches. Additionally, the SVR model is sensitive to temperature variations, whereas the DNN model is robust in this regard. Using recurrent neural networks is recommended for future ASR expansion prediction studies. Recurrent neural networks’ inherent ability to capture temporal dependencies and long-term patterns makes them well suited for analyzing sequential ultrasonic monitoring data. Overall, the results and conclusions of this study could provide insights into the capabilities and effectiveness of ML when applied to ultrasonic NDE data and help identify best practices for using ML for ultrasonic NDE of concrete material properties.

36 MATERIALS SCIENCE↗

Impact of Bone Marrow Radiation Dose on Acute Hematologic Toxicity in Cervical Cancer: Principal Component Analysis on High Dimensional Data

Purpose: To study the effects of increasing pelvic bone marrow (BM) radiation dose on acute hematologic toxicity in patients undergoing chemoradiotherapy, using a novel modeling approach to preserve the local spatial dose information. Methods and Materials: The study included 37 cervical cancer patients treated with concurrent weekly cisplatin and pelvic radiation therapy. The white blood cell count nadir during treatment was used as the indicator for acute hematologic toxicity. Pelvic BM radiation dose distributions were standardized across patients by registering the pelvic BM volumes to a common template, followed by dose remapping using deformable image registration, resulting in a dose array. Principal component (PC) analysis was applied to the dose array, and the significant eigenvectors were identified by linear regression on the PCs. The coefficients for PC regression and significant eigenvectors were represented in three dimensions to identify critical BM subregions where dose accumulation is associated with hematologic toxicity. Results: We identified five PCs associated with acute hematologic toxicity. PC analysis regression modeling explained a high proportion of the variation in acute hematologicity (adjusted R{sup 2}, 0.49). Three-dimensional rendering of a linear combination of the significant eigenvectors revealed patterns consistent with anatomical distributions of hematopoietically active BM. Conclusions: We have developed a novel approach that preserves spatial dose information to model effects of radiation dose on toxicity, which may be useful in optimizing radiation techniques to avoid critical subregions of normal tissues. Further validation of this approach in a large cohort is ongoing.

62 RADIOLOGY AND NUCLEAR MEDICINE↗

Error quantification of the Arrhenius blending rule for viscosity of hydrocarbon mixtures

Six hundred and seventy-five measurements of dynamic viscosity and density have been used to assess the prediction error of the Arrhenius blending rule for kinematic viscosity of hydrocarbon mixtures. Major trends within the data show that mixture complexity–binary to hundreds of components—and temperature are more important determinants of prediction error than differences in molecular size or hydrogen saturation between the components of the mixtures. Over the range evaluated, no correlation between prediction error and mole fractions was observed, suggesting the log of viscosity truly is linear in mole fraction, as indicated by the Arrhenius blending rule. Mixture complexity and temperature also impact molar volume and its prediction. However, a linear regression between the two model errors explains less than 20% of the observed variation, indicating that mixture viscosity and/or molar volume are not linear with respect to temperature and/or mixture complexity. Extensive discussion of the intermolecular forces and the geometric arrangement of molecules and vacancies in liquids, which ultimately determines its viscosity, is brought into context with the implicit approximations behind the Arrhenius blending rule. The complexity of this physics is not compatible with a simple algebraic correction to the model. However, sufficient data is now available to determine confidence intervals around the prediction of fuel viscosity based on its component mole fractions and viscosities. At -40°C, when all identified components are pure molecules the modeling error is 13.2% of the predicted (nominal) viscosity times the root mean square of the component mole fractions.

10 SYNTHETIC FUELS↗

In Silico Prediction of the Toxicity of Nitroaromatic Compounds: Application of Ensemble Learning QSAR Approach

In this work, a dataset of more than 200 nitroaromatic compounds is used to develop Quantitative Structure–Activity Relationship (QSAR) models for the estimation of in vivo toxicity based on 50% lethal dose to rats (LD 50 ). An initial set of 4885 molecular descriptors was generated and applied to build Support Vector Regression (SVR) models. The best two SVR models, SVR_A and SVR_B, were selected to build an Ensemble Model by means of Multiple Linear Regression (MLR). The obtained Ensemble Model showed improved performance over the base SVR models in the training set (R 2 = 0.88), validation set (R 2 = 0.95), and true external test set (R 2 = 0.92). The models were also internally validated by 5-fold cross-validation and Y-scrambling experiments, showing that the models have high levels of goodness-of-fit, robustness and predictivity. The contribution of descriptors to the toxicity in the models was assessed using the Accumulated Local Effect (ALE) technique. The proposed approach provides an important tool to assess toxicity of nitroaromatic compounds, based on the ensemble QSAR model and the structural relationship to toxicity by analyzed contribution of the involved descriptors.

54 ENVIRONMENTAL SCIENCES↗

Graphical methods for the sensitivity analysis in discriminant analysis

Similar to regression, many measures to detect influential data points in discriminant analysis have been developed. Many follow similar principles as the diagnostic measures used in linear regression in the context of discriminant analysis. Here we focus on the impact on the predicted classification posterior probability when a data point is omitted. The new method is intuitive and easily interpretative compared to existing methods. We also propose a graphical display to show the individual movement of the posterior probability of other data points when a specific data point is omitted. This enables the summaries to capture the overall pattern of the change.

97 MATHEMATICS AND COMPUTING↗

Fitting of satellite and in-situ ocean surface temperatures Results for polymode during the winter of 1977-1978

For the period considered, December 1977 through February 1978, bivariate Gaussian discriminant function cloud identification revealed that more than 93 percent of the 8-km resolution GOES infrared pixels were cloud contaminated. Cloud-free in-situ calibration points were distributed in nonrandom groups; this resulted in systematic errors when using least squares techniques. Surfaces and regression lines were least squares fitted between satellite and in-situ data; use was also made of differences and ratios. The best results were achieved with a regression in the form of the infrared radiative transfer equation; but this was no better than + or - 0.9 K. Because of extensive cloudiness, the linear regressions were seldom useful, and temperature ratios with + or - 1.3 K experimental errors best represent the applicability of GEOS data to sea surface temperatures.

Maul, G. A.↗

Application of stepwise multiple regression techniques to inversion of Nimbus 'IRIS' observations.

Exploratory studies with Nimbus-3 infrared interferometer-spectrometer (IRIS) data indicate that, in addition to temperature, such meteorological parameters as geopotential heights of pressure surfaces, tropopause pressure, and tropopause temperature can be inferred from the observed spectra with the use of simple regression equations. The technique of screening the IRIS spectral data by means of stepwise regression to obtain the best radiation predictors of meteorological parameters is validated. The simplicity of application of the technique and the simplicity of the derived linear regression equations - which contain only a few terms - suggest usefulness for this approach. Based upon the results obtained, suggestions are made for further development and exploitation of the stepwise regression analysis technique.

Ohring, G.↗

Dealing with Ion LET Uncertainties: An Application of Generalized Linear Models

Although most SEE rate estimation methods presume a fit to SEE cross section vs. LET, fitting SEE data is challenging because the data are not compatible with the assumptions of many common fitting techniques (e.g. linear regression. The difficulty of fitting such data is compounded when the LET of the ion responsible for an SEE is uncertain. We modify a Generalized Linear Model SEE data fitting methodology to accommodate uncertain LET and apply the method to the problem of backside heavy-ion SEE testing to demonstrate the utility of the method, explore the dependence of systematic errors that arise from improper treatment of LET uncertainty and develop guidelines for minimizing such systematic errors when proper treatment is not possible. Additional applications are suggested and assessed for suitability of treatment by the model.

Single-event effects↗

Standing on the Shoulders of Giants: Results From the Radiation Oncology Academic Development and Mentorship Assessment Project (ROADMAP)

Purpose: To analyze survey information regarding mentorship practices and cross-correlate the results with objective metrics of academic productivity among academic radiation oncologists at US Accreditation Council for Graduate Medical Education (ACGME)-accredited residency training programs. Methods and Materials: An institutional review board-approved survey for the Radiation Oncology Academic Development and Mentorship Assessment Project (ROADMAP) was sent to 1031 radiation oncologists employed at an ACGME-accredited residency training program and administered using an international secure web application designed exclusively to support data capture for research studies. Data collected included demographics, presence of mentorship, and the nature of specific mentoring activities. Productivity metrics, including number of publications, number of citations, h-index, and date of first publication, were collected for each survey respondent from a commercially available online database, and m-index was calculated. Results: A total of 158 academic radiation oncologists completed the survey, 96 of whom reported having an academic/scientific mentor. Faculty with a mentor had higher numbers of publications, citations, and h- and m-indices. Differences in gender and race/ethnicity were not associated with significant differences in mentorship rates, but those with a mentor were more likely to have a PhD degree and were more likely to have more time protected for research. Bivariate fit regression modeling showed a positive correlation between a mentor's h-index and their mentee's h-index (R{sup 2} = 0.16; P<.001). Linear regression also showed significant correlates of higher h-index, in addition to having a mentor (P=.001), included a longer career duration (P<.001) and fewer patients in treatment (P=.02). Conclusions: Mentorship is widely believed to be important to career development and academic productivity. These results emphasize the importance of identifying and striving to overcome potential barriers to effective mentorship.

62 RADIOLOGY AND NUCLEAR MEDICINE↗

Development of a Filtered CFD-DEM Drag Model with Multiscale Markers Using an Artificial Neural Network and Nonlinear Regression

Here, the accuracy of coarse-grained Euler-Lagrangian simulations of fluidized beds heavily depends on the mesoscale drag models to account for the influences of the unresolved sub-grid structures. Traditional filtered drag models are regressed with mesoscale markers such as voidage and slip velocities. In this research, a filtered drag was regressed with both mesoscale and macro-scale markers using fine grid Computational Fluid Dynamics - Discrete Element Method (CFD-DEM) simulations. The traditional non-linear regression method was compared with machine learning regression using an Artificial Neural Network (ANN) implemented in PyTorch and coupled with MFiX. The new drag showed higher accuracy than the Wen-Yu drag and another filtered drag derived from the two-fluid model. The nonlinear regression shows slightly better results than ANN regression in cases with similar R 2 values. The utilization of the gas inlet velocity as an additional macro-scale marker reduced the errors by up to 55.3% in the tested cases.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗

SU-E-T-30: Absorbed Doses Determined by Texture Analysis of Gafchromic EBT3 Films Using Scanning Electron Microscopy: A Feasibility Study

Purpose: The texture analysis method is useful to estimate structural features of images as color, size, and shape. The study aims to determine a dose-response curve by texture analysis of Gafchromic EBT3 film images using scanning electron microscopy (SEM). Methods: The uncoated Gafchromic EBT3 films were prepared to directly scan over the active surface layer of EBT3 film using SEM. The EBT3 films were exposed at a dose range of 0 to 10 Gy using a 6 MV photon beam. The exposed film samples were SEM-scanned at 100X, 1000X, and 3000X magnifications. The four texture features (Homogeneity, Correlation, Contrast, and Energy) were calculated based on the gray level co-occurrence matrix (GLCM) derived from the SEM images at each dose. To validate a correlation between delivered doses and texture features, an R-squared value in linear regression was tested. Results: The results showed that the Correlation index was more suitable as dose indices than the other three texture features due to higher linearity and sensitivity of the dose response curves. Further the Correlation index of 3000X magnified SEM images with 9 pixel offsets had an R-squared value of 0.964. The differences between the delivered doses and the doses measured by this method were 0.9, 1.2, 0.2, and 0.2 Gy at 5, 10, 15, and 20 Gy, respectively. Conclusion: It seems to be feasible to convert micro-scale structural features of {sub χ}t{sub χχχ}he EBT3 films to absorbed doses using the texture analysis method.

60 APPLIED LIFE SCIENCES↗

On the Observed Changes in Upper Stratospheric and Mesospheric Temperatures from UARS HALOE

Temperature versus pressure or T(p) time series from the Halogen Occultation Experiment (HALOE) on the Upper Atmosphere Research Satellite (UARS) have been extended and re-analyzed for the period of 1991-2005 and for the upper stratosphere and mesosphere in 10-degree wide latitude zones from 60S to 60N. Even though sampling from a solar occultation experiment is somewhat limited, it is shown to be quite adequate for developing both the seasonal and longer-term variations in T(p). Multiple linear regression (MLR) techniques were used in the re-analyses for the seasonal and the significant interannual, solar cycle (SC-like or decadal-scale), and linear trend terms. A simple SC-like term of 11-yr period was fitted to the time series residuals after accounting for the seasonal and interannual terms. Highly significant SC-like responses were found for both the upper mesosphere and the upper stratosphere. The phases of these SC-like terms were checked for their continuity with latitude and pressure-altitude, and in almost all cases they are directly in-phase with that of standard proxies for the solar flux variations. The analyzed, max minus min, responses at low latitudes are of order 1 K, while at middle latitudes they are as large as 3 K in the upper mesosphere. Highly significant, linear cooling trends were found at middle latitudes of the middle to upper mesosphere (about -2 K/decade), at tropical latitudes of the middle mesosphere (about -1 K/decade), and at 2 hPa (or order -1 K/decade).

Remsberg, Ellis E.↗

A Combined Al-Mg/Pb-Pb Age of the Solar System

Astrophysical models of planet formation and protoplanetary disk evolution demand precise and accurate timing of the sequence of events in the solar nebula, relative to a time t=0, usually taken to be during the short epoch of CAI (Ca-rich, Al-rich inclusion) formation. Most CAIs formed withlive26Al (mean-life τ26= 1.034 Myr [1]), with an abundance 26Al/27Al ≈ (26Al/27Al)SS= 5.23 × 10-5[2]. We adopt this as the widespread level of 26Al in the solar nebula at t=0. Assuming spatial homogeneity of 26Al, an inclusion that had less 26Al, (26Al/27Al)0, formed a time Δt26= τ26ln[(26Al/27Al)SS/ (26Al/27Al)0] after t=0.These ages are typical precise to within ±0.1 Myr. Igneous bulk meteorites and inclusions can be relatively dated by the Al-Mg chronometer, but only ifΔt26<6 Myr. The Pb-Pb system is useful as a longer relative chronometer. It yields absolute ages tPb using 207Pb/206Pb, 206Pb/204Pb, and 238U/235U ratios measured indifferent portions of a sample, assuming certain half-lives [4]. These absolute ages are uncertain to within ±9 Myr due to uncertainties in the 235U half-life[3], but times of formation ΔtPb= tCAI–tPb relative to t=0, are more precise(±0.5Myr),iftCAI can be found. Here, tCAI means the Pb-Pb age that would be measured in CAIs using the half-lives the community typically uses, if they achieved isotopic closure at t=0. Unfortunately, direct Pb-Pb dating of CAIs has not definitively determined tCAI. Based on four CAIs with canonical (26Al/27Al)0,[5,6] found tPb= 4567.30 ± 0.16 Myr. No other CAI ages with measured 238U/235U have been reported in the refereed literature, but there are hints of other CAIs with ages tPb= 4568.0 ± 0.2 Myr [7] and tPb= 4568.3 ± 0.2 Myr [8].It is unclear whether anyof these igneous type B CAIs isotopically closed at t=0 or represents tCAI. Instead of measurements, we advocate finding tCAI by minimizing the discrepancies between the Al-Mg and Pb-Pb chronometers. Assuming Δt26=ΔtPb, we find the implied t’CAI= tPb+Δt26, then define t*CAIas the weighted mean of the t’CAI. t*CAIis the best guess for the Pb-Pb age of t=0; the assumption of homogeneity is justified if the t’CAI cluster within errors around t*CAI. This statistical approach is similar to, but improves on, that of[9]. We find t*CAI= 4568.73 ± 0.16 Myr. Below we discuss our methodology and the implications of this age for CAIs, 1.4 Myr older than the reported and typically used age 4567.30±0.16 Myr. Methods: We base our estimate of t*CAIon five achondrites for which published (26Al/27Al)0and Pb-Pb ages exist: the quenched angrites D’Orbigny, Sahara 99555 (SAH 99555), and Northwest Africa (NWA) 1670; the pseudo-eucrite Asuka 881394; and the inner disk achondrite. All are “NC” (non-carbonaceous) achondrites that likely cooled quickly enough that the Al-Mg and Pb-Pb systems achieved isotopic closure simultaneously. We also considered the “CC” (carbonaceous chondrite-like) achondrites NWA 2796 and NWA 6704, butdo not include them in our fit. Al-Mg and Pb-Pb seem not to have closed simultaneously, possibly because formation in the outer disk from volatile-rich composition led to slower cooling. Of the 8 chondrules from NWA 5697 measured by [20,21], we also consider the 4 for which 238U/235U was measured: 2-C1, 5-C2, 3-C5, 11-C1.Depending on their post-formation thermal histories, the Al-Mg and Pb-Pb systems in chondrules may or may not have closed simultaneously. Table 1: (26Al/27Al)0, Pb-Pb ages of selected samplesSample(26Al/27Al)0/ 10-6RefPb-PbRefD’Orbigny3.98±0.15104563.43±0.19♮10-12SAH 995553.64±0.18104563.88±0.2712NWA16705.92±0.59104564.39±0.24*10Asuka 88139413.1±0.5613-154564.98±0.1715NWA 73253.03±0.14164563.4±2.616NWA 27963.94±0.16174562.89±0.5917NWA 67043.03±0.14184562.76±0.26192-C17.56±1.53204567.57±0.56*215-C27.04±1.51204567.54±0.52*213-C58.85±1.83204566.20±0.63*2111-C15.55±1.84204565.84±0.72*21*regression based on one subset of data points ♮weighted mean of two datasets Pb-Pb ages are proportional to the intercept of the line formed by linear regression of 207Pb/206Pb vs. 206Pb/204Pb data from various washes, leachates and residues of acid dissolution of a sample. Because contamination by terrestrial or primordial Pb is pervasive, some fractions must be excluded from regressions to ensure a fit with acceptable mean squares weighted deviation (MSWD). Usually points are excluded based on low [Pb], or low 206Pb/204Pb ratio(low radiogenic component), with single outliers identified [11,12,15,16,17]. In the starred examples (Table 1)and the case of 3 CAI Pb-Pb ages [5], up to half the points were excluded solely because did not fit a pre-determined line. This approach is vulnerable to confirmation bias and produces fits with low MSWD and too-low Pb-Pb age uncertainty. Regressing the same data points as [10], were produce the Pb-Pb age of NWA 1670 of 4564.39±0.24 Myr. But selecting other combinations of data points, other, equally valid, isochrons yield ages from 4563.77±0.21 Myr to 4564.64±0.23 Myr. Similar arguments apply to the Pb-Pb isochrons built by [21] for chondrules 2-C1 (we find 4567.33±0.44 to 4567.85±0.46 Myr), 5-C2 (4566.84±0.53 to 4567.70±0.44 Myr), 3-C5 (4565.84±0.54to 4567.04±0.54)and 11-C1 (4565.36±0.51 to 4565.74±0.45 Myr). Our adopted ages for these and NWA 1670 are listed in Table 2.Table 2. tCAI estimated from various components, using our regressions for the chondrules & NWA 1670.SampleΔt26(Myr)tPb(Myr)t’CAI(Myr)D’Orbigny5.05±0.044563.43±0.194568.48±0.19SAH 995555.14±0.054563.88±0.274569.02±0.27NWA16704.64±0.104564.21±0.634568.85±0.67Asuka 8813943.81±0.044564.98±0.174568.79±0.17NWA 73255.33±0.054563.4±2.64568.7±2.6NWA 27965.06±0.044562.89±0.594567.95±0.59NWA 67045.29±0.134562.76±0.264568.05±0.292-C12.00±0.214567.59±0.704569.59±0.725-C22.07±0.224567.23±0.914569.30±0.933-C51.84±0.214566.44±1.124568.28±1.1411-C12.32±0.344565.52±0.664567.84±0.73achondrite4568.72±0.16chondrules4568.76±0.58combined4568.73±0.16A weighted average of the five NC achondrites(or just D’Orbigny, SAH 99555 and Asuka 881394)yields t*CAI= 4568.72 ± 0.16Myr. All are consistent with this value to within 1.8σ, and MSWD=1.5. Including the 4 U-corrected chondrules, t*CAI= 4568.73± 0.16Myrwith MSWD=1.66, which is statistically significant. All chondrules and NC achondrites are consistent with this to within 1.8σ, (Figure 1).Figure 1. Al-Mg formation times after t=0 vs. Pb-Pb ages. The five NC achondrites and four chondrules are consistent with a Pb-Pb age of t=0 of 4568.7 Myr. Discussion: The data from achondrites and chondrules are consistent with a single Pb-Pb age at t=0, justifying the assumption of 26Al homogeneity. The age, 4568.7 Myr, is ≈1.4 Myr older than the commonly accepted Pb-Pb age of CAIs that formed with canonical 26Al/27Al at t=0 [3]. Others have interpreted the discrepancy to signify 26Al heterogeneity in the CAI-forming region[5,21]. We suggest instead that CAIs were exposed to transient heating events that reset the Pb-Pb system without disturbing the Al-Mg system. Notably, chondrules typically experienced transient heating at these times in the nebula [22]. If so, direct measurements of CAIs will not yield as reliable a Pb-Pb age of t=0 as statistical approaches like this and that of [9].References:[1] Auer et al. 2009. [2] Jacobsen, B et al. 2008, EPSL 272, 353-364. [3] Tissot, Fet al. 2017, GCA 213, 593-617.[4] Villa, I et al. 2016, GCA 172, 387-392.[5] Amelin, Y et al. 2010, EPSL 300, 343-350.[6] Connelly, Jet al. 2012, Science 338, 651.[7] Bouvier, Aet al. 2011, LPICo 1639, 9054. [8] Bouvier, Aand Wadhwa, M2010, Nat Geosci 3, 637-641.[9] Nyquist, Let al. 2009, GCA 73, 5115-5136.[10] Schiller et al. 2015. [11] Wadhwa & Brennecka 2012. [12] Tissot et al. 2017. [13] Nyquist et al. 2003. [14] Wadhwa et al. 2009. [15] Wimpenny et al. 2019, GCA 244, 478-501.[16] Koefoed et al. 2016, GCA 183, 31-45. [17] Bouvier, A et al. 2011, GCA 75, 5310-5323. [18] Sanborn, Met al. 2019, GCA 245, 577-596. [19] Amelin, Yet al. 2019, GCA 245, 628-642.[20] Bollard, Jet al. 2017, Sci Adv 3 ,e1700407. [21] Bollard, Jet al. 2019, GCA 260, 62-83.[22] Villeneuve, J et al. 2009, Science 325, 985

S. J. Desch↗

Evaluating the Trustworthiness of Explainable Artificial Intelligence (XAI) Methods Applied to Regression Predictions of Arctic Sea Ice Motion

Abstract Recent advances in explainable artificial intelligence (XAI) methods show promise for understanding predictions made by machine learning (ML) models. XAI explains how the input features are relevant or important for the model predictions. We train linear regression (LR) and convolutional neural network (CNN) models to make 1-day predictions of sea ice velocity in the Arctic from inputs of present-day wind velocity and previous-day ice velocity and concentration. We apply XAI methods to the CNN and compare explanations to variance explained by LR. We confirm the feasibility of using a novel XAI method [i.e., global layerwise relevance propagation (LRP)] to understand ML model predictions of sea ice motion by comparing it to established techniques. We investigate a suite of linear, perturbation-based, and propagation-based XAI methods in both local and global forms. Outputs from different explainability methods are generally consistent in showing that wind speed is the input feature with the highest contribution to ML predictions of ice motion, and we discuss inconsistencies in the spatial variability of the explanations. Additionally, we show that the CNN relies on both linear and nonlinear relationships between the inputs and uses nonlocal information to make predictions. LRP shows that wind speed over land is highly relevant for predicting ice motion offshore. This provides a framework to show how knowledge of environmental variables (i.e., wind) on land could be useful for predicting other properties (i.e., sea ice velocity) elsewhere. Significance Statement Explainable artificial intelligence (XAI) is useful for understanding predictions made by machine learning models. Our research establishes trustability in a novel implementation of an explainable AI method known as layerwise relevance propagation for Earth science applications. To do this, we provide a comparative evaluation of a suite of explainable AI methods applied to machine learning models that make 1-day predictions of Arctic sea ice velocity. We use explainable AI outputs to understand how the input features are used by the machine learning to predict ice motion. Additionally, we show that a convolutional neural network uses nonlinear and nonlocal information in making its predictions. We take advantage of the nonlocality to investigate the extent to which knowledge of wind on land is useful for predicting sea ice velocity elsewhere.

Hoffman, Lauren [Scripps Institution of Oceanograp↗

Trends and Solar Cycle Effects in Temperature Versus Altitude From the Halogen Occultation Experiment for the Mesosphere and Upper Stratosphere

Fourteen-year time series of mesospheric and upper stratospheric temperatures from the Halogen Occultation Experiment (HALOE) are analyzed and reported. The data have been binned according to ten-degree wide latitude zones from 40S to 40N and at 10 altitudes from 43 to 80 km-a total of 90 separate time series. Multiple linear regression (MLR) analysis techniques have been applied to those time series. This study focuses on resolving their 11-yr solar cycle (or SC-like) responses and their linear trend terms. Findings for T(z) from HALOE are compared directly with published results from ground-based Rayleigh lidar and rocketsonde measurements. SC-like responses from HALOE compare well with those from lidar station data at low latitudes. The cooling trends from HALOE also agree reasonably well with those from the lidar data for the concurrent decade. Cooling trends of the lower mesosphere from HALOE are not as large as those from rocketsondes and from lidar station time series of the previous two decades, presumably because the changes in the upper stratospheric ozone were near zero during the HALOE time period and did not affect those trends.

Remsberg, Ellis E.↗

Forecasting Megaelectron-Volt Electrons Inside Earth's Outer Radiation Belt: PreMevE 2.0 Based on Supervised Machine Learning Algorithms

We present the recent progress in upgrading a predictive model for megaelectron-volt (MeV) electrons inside the Earth's outer Van Allen belt. This updated model, called PreMevE 2.0, provides improved forecasts, particularly at outer L-shells, by adding upstream solar wind speeds to the model's input parameter list that originally includes precipitating electrons observed at low Earth orbits and MeV electron fluxes in situ measured by a geosynchronous satellite. Furthermore, based on several kinds of linear and artificial neural networks algorithms, a list of models was constructed, trained, validated, and tested with 42-month MeV electron observations from Van Allen Probes. Out-of-sample test results from these models show that, with optimized model hyperparameters and input parameter combinations, the top performer from each category of models has the similar capability of making reliable 1-day (2-day) forecasts of 1-MeV electron flux distributions with performance efficiency values ~0.87 (~0.82) averaged over the L-shell range of 2.8–6.6, significantly outperforming the previous version of PreMevE particularly at L-shells > ~4.5. Interestingly, the linear regression model is often the most successful when compared to other models, which suggests the relationship between dynamics of trapped 1-MeV electrons and precipitating electrons is dominated by linear components. Results also show that PreMevE 2.0 can reasonably well predict the onsets of MeV electron events in 2-day forecasts. PreMevE 2.0 is designed to be driven by observations from longstanding space infrastructure to make high-fidelity forecasts for MeV electrons, thus can be an invaluable space weather forecasting tool for the future.

79 ASTRONOMY AND ASTROPHYSICS↗

Laboratory Headphone Studies of Human Response to Low-Amplitude Sonic Booms and Rattle Heard Indoors

Human response to sonic booms heard indoors is affected by the generation of contact-induced rattle noise. The annoyance caused by sonic boom-induced rattle noise was studied in a series of psychoacoustics tests. Stimuli were divided into three categories and presented in three different studies: isolated rattles at the same calculated Perceived Level (PL), sonic booms combined with rattles with the mixed sound at a single PL, and sonic booms combined with rattles with the mixed sound at three different PL. Subjects listened to sounds over headphones and were asked to report their annoyance. Annoyance to different rattles was shown to vary significantly according to rattle object size. In addition, the combination of low-amplitude sonic booms and rattles can be more annoying than the sonic boom alone. Correlations and regression analyses for the combined sonic boom and rattle sounds identified the Moore and Glasberg Stationary Loudness (MGSL) metric as a primary predictor of annoyance for the tested sounds. Multiple linear regression models were developed to describe annoyance to the tested sounds, and simplifications for applicability to a wider range of sounds are presented.

Loubeau, Alexandra↗

Objective Lightning Forecasting at Kennedy Space Center/Cape Canaveral Air Force Station using Cloud-to-Ground Lightning Surveillance System Data

The 45th Weather Squadron (45 WS) forecasters at Cape Canaveral Air Force Station (CCAFS) in Florida include a probability of thunderstorm occurrence in their daily morning briefings. This information is used by personnel involved in determining the possibility of violating Launch Commit Criteria, evaluating Flight Rules for the Space Shuttle, and daily planning for ground operation activities on Kennedy Space Center (KSC)/CCAFS. Much of the current lightning probability forecast is based on a subjective analysis of model and observational data. The forecasters requested that a lightning probability forecast tool based on statistical analysis of historical warm-season (May - September) data be developed in order to increase the objectivity of the daily thunderstorm probability forecast. The tool is a set of statistical lightning forecast equations that provide a lightning occurrence probability for the day by 1100 UTC (0700 EDT) during the warm season. This study used 15 years (1989-2003) of warm season data to develop the objective forecast equations. The local CCAFS 1000 UTC sounding was used to calculate stability parameters for equation predictors. The Cloud-to-Ground Lightning Surveillance System (CGLSS) data were used to determine lightning occurrence for each day. The CGLSS data have been found to be more reliable indicators of lightning in the area than surface observations through local informal analyses. This work was based on the results from two earlier research projects. Everitt (1999) used surface observations and rawinsonde data to develop logistic regression equations that forecast the daily thunderstorm probability at CCAFS. The Everitt (1999) equations showed an improvement in skill over the Neumann-Pfeffer thunderstorm index (Neumann 1971), which uses multiple linear regression, and also persistence and climatology forecasts. Lericos et al. (2002) developed lightning distributions over the Florida peninsula based on specific flow regimes. The flow regimes were inferred from the average wind direction in the 1000-700 mb layer at Miami (MIA), Tampa (TBW), and Jacksonville (JAX), Florida, and the lightning data were from the National Lightning Detection Network. The results suggested that the daily flow regime may be an important predictor of lightning occurrence on KSC/CCAFS.

Lambert, Winifred↗