SEARCH · Engineering Papers

Results for “Input”

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.

SourceResearch connectionFrom yearThrough year

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

At least 1,675 records · Page 93

All sources NASA NTRS DOE OSTI

DOE OSTI2021

Turn-key constrained parameter space exploration for particle accelerators using Bayesian active learning

Abstract Particle accelerators are invaluable discovery engines in the chemical, biological and physical sciences. Characterization of the accelerated beam response to accelerator input parameters is often the first step when conducting accelerator-based experiments. Currently used techniques for characterization, such as grid-like parameter sampling scans, become impractical when extended to higher dimensional input spaces, when complicated measurement constraints are present, or prior information known about the beam response is scarce. Here in this work, we describe an adaptation of the popular Bayesian optimization algorithm, which enables a turn-key exploration of input parameter spaces. Our algorithm replaces the need for parameter scans while minimizing prior information needed about the measurement’s behavior and associated measurement constraints. We experimentally demonstrate that our algorithm autonomously conducts an adaptive, multi-parameter exploration of input parameter space, potentially orders of magnitude faster than conventional grid-like parameter scans, while making highly constrained, single-shot beam phase-space measurements and accounts for costs associated with changing input parameters. In addition to applications in accelerator-based scientific experiments, this algorithm addresses challenges shared by many scientific disciplines, and is thus applicable to autonomously conducting experiments over a broad range of research topics.

43 PARTICLE ACCELERATORS↗

DOE OSTI2024

Network-level encoding of local neurotransmitters in cortical astrocytes

Abstract Astrocytes, the most abundant non-neuronal cell type in the mammalian brain, are crucial circuit components that respond to and modulate neuronal activity through calcium (Ca 2+ ) signalling 1–7 . Astrocyte Ca 2+ activity is highly heterogeneous and occurs across multiple spatiotemporal scales—from fast, subcellular activity 3,4 to slow, synchronized activity across connected astrocyte networks 8–10 —to influence many processes 5,7,11 . However, the inputs that drive astrocyte network dynamics remain unclear. Here we used ex vivo and in vivo two-photon astrocyte imaging while mimicking neuronal neurotransmitter inputs at multiple spatiotemporal scales. We find that brief, subcellular inputs of GABA and glutamate lead to widespread, long-lasting astrocyte Ca 2+ responses beyond an individual stimulated cell. Further, we find that a key subset of Ca 2+ activity—propagative activity—differentiates astrocyte network responses to these two main neurotransmitters, and may influence responses to future inputs. Together, our results demonstrate that local, transient neurotransmitter inputs are encoded by broad cortical astrocyte networks over a minutes-long time course, contributing to accumulating evidence that substantial astrocyte–neuron communication occurs across slow, network-level spatiotemporal scales 12–14 . These findings will enable future studies to investigate the link between specific astrocyte Ca 2+ activity and specific functional outputs, which could build a consistent framework for astrocytic modulation of neuronal activity.

Science & Technology - Other Topics↗

DOE OSTI2021

An adaptive approach to machine learning for compact particle accelerators

Abstract Machine learning (ML) tools are able to learn relationships between the inputs and outputs of large complex systems directly from data. However, for time-varying systems, the predictive capabilities of ML tools degrade if the systems are no longer accurately represented by the data with which the ML models were trained. For complex systems, re-training is only possible if the changes are slow relative to the rate at which large numbers of new input-output training data can be non-invasively recorded. In this work, we present an approach to deep learning for time-varying systems that does not require re-training, but uses instead an adaptive feedback in the architecture of deep convolutional neural networks (CNN). The feedback is based only on available system output measurements and is applied in the encoded low-dimensional dense layers of the encoder-decoder CNNs. First, we develop an inverse model of a complex accelerator system to map output beam measurements to input beam distributions, while both the accelerator components and the unknown input beam distribution vary rapidly with time. We then demonstrate our method on experimental measurements of the input and output beam distributions of the HiRES ultra-fast electron diffraction (UED) beam line at Lawrence Berkeley National Laboratory, and showcase its ability for automatic tracking of the time varying photocathode quantum efficiency map. Our method can be successfully used to aid both physics and ML-based surrogate online models to provide non-invasive beam diagnostics.

97 MATHEMATICS AND COMPUTING↗

DOE OSTI2022

Dynamics of photosynthetic light harvesting systems interacting with N-photon Fock states

Here, we develop a method to simulate the excitonic dynamics of realistic photosynthetic light harvesting systems, including non-Markovian coupling to phonon degrees of freedom, under excitation by N-photon Fock state pulses. This method combines the input–output and the hierarchical equations of motion formalisms into a double hierarchy of density matrix equations. We show analytically that under weak field excitation relevant to natural photosynthesis conditions, an N-photon Fock state input and a corresponding coherent state input give rise to equal density matrices in the excited manifold. However, an N-photon Fock state input induces no off-diagonal coherence between the ground and excited subspaces, in contrast with the coherences created by a coherent state input. We derive expressions for the probability to absorb a single Fock state photon with or without the influence of phonons. For short pulses (or, equivalently, wide bandwidth pulses), we show that the absorption probability has a universal behavior that depends only upon a system-dependent effective energy spread parameter Δ and an exciton–light coupling constant Γ. This holds for a broad range of chromophore systems and for a variety of pulse shapes. We also analyze the absorption probability in the opposite long pulse (narrow bandwidth) regime. We then derive an expression for the long time emission rate in the presence of phonons and use it to study the difference between collective vs independent emission. Finally, we present a numerical simulation for the LHCII monomer (14-mer) system under single photon excitation that illustrates the use of the double hierarchy equations.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗

DOE OSTI2023

Enhancing Monte Carlo Workflows for Nuclear Reactor Analysis with Metamodel-Driven Modeling

Monte Carlo codes are essential components of many reactor physics simulation workflows as high-fidelity continuous-energy neutron transport solvers. Among Monte Carlo radiation transport codes, MCNP is particularly notable due to its diverse simulation capabilities, large user base, and long validation history. Despite being a powerful simulation tool, MCNP provides limited capabilities to allow automated execution, model transformation, or support for user-defined logic and abstractions that limit its compatibility with modern workflows. Here, to better integrate MCNP into a modern scientific workflow, we have developed an intuitive yet full-featured MCNP Application Program Interface (API) in Python, named MCNPy, which provides a specialized set of classes for MCNP input development. Moreover, to guarantee that our reading, writing, and modeling capabilities remain self-consistent (and to render the huge scope of the MCNP API manageable), we have adopted a strategy of model-driven software development in which a generalized model of the MCNP input format has been created. From this generalized model, or “metamodel,” problem-specific implementations such as an engine for input validation or a codebase for programmatic operations may be automatically generated. Since MCNPy primarily acts as a Python front-end to the underlying Java API that directly interfaces with the metamodel, it is intrinsically linked to the metamodel and thus remains maintainable. With MCNPy, users can programmatically read, write, and modify any syntactically valid MCNP input file regardless of its origin. These capabilities allow users to automate complicated tasks like design optimization and model translation for nuclear systems. As examples, this work demonstrates the use of MCNPy to find the critical radius of a plutonium sphere and to translate a 9000+ line MCNP input file into a corresponding OpenMC model.

22 GENERAL STUDIES OF NUCLEAR REACTORS↗

DOE OSTI2015

Identifying the location of the OMP separatrix in DIII-D using power accounting

In order to identify reliable scalings for the scrape-off layer (SOL) power width it is necessary to know the location of the separatrix in divertor tokamaks as accurately as possible, specifically its location at the outside midplane (OMP) the standard reference location. Two methods are described which use power accounting to improve the accuracy of identifying the location of the OMP separatrix. The first uses the infrared-measured deposited power profile at the outer target as the primary input, the ' $P_{{\rm SOL}}^{{\rm exhaust}}$ method'. The second uses the measured power input to the SOL, obtained by subtracting the power radiated from inside the separatrix from the total heating power, the ' $P_{{\rm SOL}}^{{\rm input}}$ method'. Furthermore, these two power accounting methods are illustrated with the examples of 21 H-mode DIII-D discharges. High spatial resolution Thomson scattering measured profiles of ne and Te for the main SOL near the OMP are also used as primary input to the analysis; only between-edge localized mode data are used here. The Thomson profiles are used to calculate the electron parallel conducted heat flux profiles which are then matched to the measured $P_{{\rm SOL}}^{{\rm exhaust}}$ and $P_{{\rm SOL}}^{{\rm input}}$ by adjusting the location of the OMP separatrix relative to that of the Thomson data. For these attached discharges, it is found that the values of $R_{{\rm sep}}^{{\rm omp}}$ given by the two power accounting methods agree to within ~1 mm of each other and also to within ~1 mm of the values given by the 'standard DIII-D method' described by Porter et al (1998 Phys. Plasmas 5 1410). Lastly, the shifted $R_{{\rm sep}}^{{\rm omp}}$ results in only modest changes to the values of ne and Te at the OMP separatrix relative to the 'standard' values, increasing $n_{{\rm e}}^{{\rm sep}}$ by 8% and $T_{{\rm e}}^{{\rm sep}}$ by 20%.

70 PLASMA PHYSICS AND FUSION TECHNOLOGY↗

DOE OSTI2016

SWAP-Assembler 2: Optimization of De Novo Genome Assembler at Large Scale

In this paper, we analyze and optimize the most time-consuming steps of the SWAP-Assembler, a parallel genome assembler, so that it can scale to a large number of cores for huge genomes with the size of sequencing data ranging from terabyes to petabytes. According to the performance analysis results, the most time-consuming steps are input parallelization, k-mer graph construction, and graph simplification (edge merging). For the input parallelization, the input data is divided into virtual fragments with nearly equal size, and the start position and end position of each fragment are automatically separated at the beginning of the reads. In k-mer graph construction, in order to improve the communication efficiency, the message size is kept constant between any two processes by proportionally increasing the number of nucleotides to the number of processes in the input parallelization step for each round. The memory usage is also decreased because only a small part of the input data is processed in each round. With graph simplification, the communication protocol reduces the number of communication loops from four to two loops and decreases the idle communication time. The optimized assembler is denoted as SWAP-Assembler 2 (SWAP2). In our experiments using a 1000 Genomes project dataset of 4 terabytes (the largest dataset ever used for assembling) on the supercomputer Mira, the results show that SWAP2 scales to 131,072 cores with an efficiency of 40%. We also compared our work with both the HipMER assembler and the SWAP-Assembler. On the Yanhuang dataset of 300 gigabytes, SWAP2 shows a 3X speedup and 4X better scalability compared with the HipMer assembler and is 45 times faster than the SWAP-Assembler. The SWAP2 software is available at https://sourceforge.net/projects/swapassembler.

Meng, Jintao↗

DOE OSTI2018

Correlation between EVs and Other Loads in Probabilistic Load Flow for Distribution Systems

To date, most probabilistic studies have assumed independence between the various loads and other inputs on the distribution system. It is well know however, that this is typically not the case. The work presented in the paper expands upon a previous unbalanced three-phase probabilistic load flow study on a residential feeder with a 50 % penetration of electric vehicles, under three different charging schemes where different correlation scenarios between the load flow inputs are considered. The data for the probabilistic load flow study are derived from a voltage dependent time-variant deterministic load flow. Here the deterministic load flow data are examined further to gain a better understanding of the correlation between the customer and electric vehicle charging loads, and how different charging schemes may impact it further. Four different correlation scenarios are then created and details on how the probabilistic load flow inputs are created is also presented; demonstrating the need for better analysis on the correlation between system inputs. This paper hopes to open the discussion and start the push for future research into understanding the correlation and effects of various system inputs on the probabilistic nature of distribution systems, and how to apply this knowledge to future studies.

Melhorn, Alexander C.↗

DOE OSTI2019

Plant roots stimulate the decomposition of complex, but not simple, soil carbon

Roots release carbon into soil and can alleviate energy limitation of microbial organic matter decomposition. We know little about the effects of roots on microbial decomposition of different organic matter substrates, despite the importance for soil carbon stocks and turnover. Through implementing root-microbe interactions, the Carbon, Organisms, Rhizosphere, and Protection in the Soil Environment (CORPSE) model was previously shown to represent dynamics of total soil carbon in temperate forest field experiments. However, the model permits alternative hypotheses concerning microbial-substrate affinity. We investigated how root inputs affect decomposition of soil organic carbon (SOC) with variable decomposability. We simulated SOC stocks in CORPSE and compared microbial degradation of two substrates types with varying root-microbe interactions under two alternative hypotheses that varied in microbial-substrate affinity. We compared our modeled hypotheses to a forest field experiment where we quantified decomposition of isotopically-labeled starch and leaf tissues in soils with manipulated root access to microbes. Here, we tested the hypothesis that decomposition of leaves would be more sensitive to root inputs than decomposition of starch, corresponding to the alternative model hypothesis. In the field study, leaf decomposition increased with root density while starch decomposition was unchanged by root density. Microbial biomass and enzyme activity consistently increased with root inputs in CORPSE and the field study. Our field experiment supported the CORPSE simulations with high microbial-substrate affinity. Roots stimulated microbial growth and enzyme production, which increased degradation of more complex substrates such as leaf tissues. Substrates that were easily decomposed, such as starch, may already be degrading at a maximum rate in the absence of rhizosphere influence because their decomposition rate was unchanged by root inputs. We found that the degree to which roots stimulate microbial decomposition depends on the substrate being decomposed, and that root-microbe interactions influenced SOC stocks in both our model and field experiment. Environmental changes that alter root-microbe interactions could, therefore, alter soil C stocks and biogeochemical cycling, and models of these interactions should incorporate differential influence of rhizosphere inputs on different substrates.

59 BASIC BIOLOGICAL SCIENCES↗

DOE OSTI2021

Leveraging observed soil heterotrophic respiration fluxes as a novel constraint on global-scale models

Microbially-explicit models may improve understanding and projections of carbon dynamics in response to future climate change, but their fidelity in simulating global-scale soil heterotrophic respiration (RH), a stringent test for soil biogeochemical models, has never been evaluated. We used statistical global RH products, as well as 7,821 daily site-scale RH measurements, to evaluate the spatio-temporal performance of one first-order decay model (CASA-CNP) and two microbially-explicit biogeochemical models (CORPSE and MIMICS) that were forced by two different climate datasets. CORPSE and MIMICS did not provide any measurable performance improvement; instead, the models were highly sensitive to the meteorological input data used to drive them. Spatial RH variability was generally well simulated except in the northern middle latitudes (~50°N) and arid regions; models captured the seasonal variability of RH well, but showed more divergence in tropic and arctic regions. Our results demonstrate that the next generation of biogeochemical models shows promise, but also needs to be improved for realistic spatio-temporal variability of RH. Finally, we emphasize the importance of net primary production, soil moisture, and soil temperature inputs, and that jointly evaluating soil models for their spatial (global scale) and temporal (site scale) performance provides crucial benchmarks for improving biogeochemical models.

Jian, Jinshi↗

DOE OSTI2026

Drought Reduces Formation, but Enhances Persistence, of Mineral‐Associated Organic Matter in a Grassland Soil

Drought effects are pervasive in terrestrial ecosystems, yet there is limited understanding of how drought impacts the transformation of plant carbon (C) inputs to mineral-associated organic matter (MAOM)—the largest and slowest-cycling pool of soil organic carbon (SOC). In a 12-week 13 C-CO 2 greenhouse labeling experiment, we tracked the formation of MAOM derived from the two dominant sources of plant C input to the mineral soil—living root inputs ( 13 C-rhizodeposits) and decaying root inputs ( 13 C-root detritus)—under normal moisture and droughted conditions in a semiarid grassland soil. At the end of the 12-week period, we also measured the persistence of 13 C-MAOM formed from rhizodeposits versus root detritus via a subsequent persistence assay. Drought reduced the formation of MAOM derived from living roots by decreasing rhizodeposits, reducing microbial growth rates, and altering the composition of organic matter, lipids, and metabolites. Drought initially delayed the formation of MAOM derived from root detritus by slowing the early stages of root litter decomposition (week 4–8), but did not decrease total MAOM formation by the end of the 12-week period. Notably, drought enhanced the persistence of MAOM derived from root detritus, but did not influence the persistence of MAOM derived from rhizodeposits. Our results provide some of the first direct evidence that drought can reduce the formation of MAOM in a grassland soil, but may enhance its persistence, based on the source of plant input from which MAOM is derived.

13C- labeling↗

DOE OSTI2017

Cover crop root contributions to soil carbon in a no‐till corn bioenergy cropping system

Abstract Crop residues are potential biofuel feedstocks, but residue removal may reduce soil carbon (C). The inclusion of a cover crop in a corn bioenergy system could provide additional biomass, mitigating the negative effects of residue removal by adding to stable soil C pools. In a no‐till continuous corn bioenergy system in the northern US Corn Belt, we used 13 CO 2 pulse labeling to trace plant C from a winter rye ( Secale cereale ) cover crop into different soil C pools for 2 years following rye cover crop termination. Corn stover left as residue (30% of total stover) contributed 66, corn roots 57, rye shoots 61, rye roots 50, and rye rhizodeposits 25 g C m −2 to soil. Five months following cover crop termination, belowground cover crop inputs were three times more likely to remain in soil C pools than were aboveground inputs, and much of the root‐derived C was in mineral‐associated soil fractions. After 2 years, both above‐ and belowground inputs had declined substantially, indicating that the majority of both root and shoot inputs are eventually mineralized. Our results underscore the importance of cover crop roots vs. shoots and the importance of cover crop rhizodeposition (33% of total belowground cover crop C inputs) as a source of soil C. However, the eventual loss of most cover crop C from these soils indicates that cover crops will likely need to be included every year in rotations to accumulate soil C.

13CO2↗

DOE OSTI2021

Future roots for future soils

Abstract Mechanical impedance constrains root growth in most soils. Crop cultivation changed the impedance characteristics of native soils, through topsoil erosion, loss of organic matter, disruption of soil structure and loss of biopores. Increasing adoption of Conservation Agriculture in high‐input agroecosystems is returning cultivated soils to the soil impedance characteristics of native soils, but in the low‐input agroecosystems characteristic of developing nations, ongoing soil degradation is generating more challenging environments for root growth. We propose that root phenotypes have evolved to adapt to the altered impedance characteristics of cultivated soil during crop domestication. The diverging trajectories of soils under Conservation Agriculture and low‐input agroecosystems have implications for strategies to develop crops to meet global needs under climate change. We present several root ideotypes as breeding targets under the impedance regimes of both high‐input and low‐input agroecosystems, as well as a set of root phenotypes that should be useful in both scenarios. We argue that a ‘whole plant in whole soil’ perspective will be useful in guiding the development of future crops for future soils.

Lynch, Jonathan P.↗

DOE OSTI2021

Unexpected Scaling of Peak Acceleration for a Yielding Mass-Spring System Subjected to a Triangular Base Acceleration Pulse

The use of bounding scenarios is a common practice that greatly simplifies the design and qualification of structures. However, this approach implicitly assumes that the quantities of interest increase monotonically with the input to the structure, which is not necessarily true for nonlinear structures. This paper surveys the literature for observations of nonmonotonic behavior of nonlinear systems and finds such observations in both the earthquake engineering and applied mechanics literature. Numerical simulations of a single degree-of-freedom mass-spring system with an elastic–plastic spring subjected to a triangular base acceleration pulse are then presented, and it is shown that the relative acceleration of this system scales nonmonotonically with the input magnitude in some cases. The equation of motion for this system is solved symbolically and an approximate expression for the relative acceleration is developed, which qualitatively agrees with the nonmonotonic behavior seen in the numerical results. The nonmonotonicity is investigated and found to be a result of dynamics excited by the discontinuous derivative of the base acceleration pulse, the magnitude of which scales nonmonotonically with the input magnitude due to the fact that the first yield of the spring occurs earlier as the input magnitude is increased. The relevance of this finding within the context of defining bounding scenarios is discussed, and it is recommended that modeling be used to perform a survey of the full range of possible inputs prior to defining bounding scenarios.

42 ENGINEERING↗

DOE OSTI2016

SU-F-J-105: Towards a Novel Treatment Planning Pipeline Delivering Pareto- Optimal Plans While Enabling Inter- and Intrafraction Plan Adaptation

Purpose: To develop a new IMRT treatment planning methodology suitable for the new generation of MR-linear accelerator machines. The pipeline is able to deliver Pareto-optimal plans and can be utilized for conventional treatments as well as for inter- and intrafraction plan adaptation based on real-time MR-data. Methods: A Pareto-optimal plan is generated using the automated multicriterial optimization approach Erasmus-iCycle. The resulting dose distribution is used as input to the second part of the pipeline, an iterative process which generates deliverable segments that target the latest anatomical state and gradually converges to the prescribed dose. This process continues until a certain percentage of the dose has been delivered. Under a conventional treatment, a Segment Weight Optimization (SWO) is then performed to ensure convergence to the prescribed dose. In the case of inter- and intrafraction adaptation, post-processing steps like SWO cannot be employed due to the changing anatomy. This is instead addressed by transferring the missing/excess dose to the input of the subsequent fraction. In this work, the resulting plans were delivered on a Delta4 phantom as a final Quality Assurance test. Results: A conventional static SWO IMRT plan was generated for two prostate cases. The sequencer faithfully reproduced the input dose for all volumes of interest. For the two cases the mean relative dose difference of the PTV between the ideal input and sequenced dose was 0.1% and −0.02% respectively. Both plans were delivered on a Delta4 phantom and passed the clinical Quality Assurance procedures by achieving 100% pass rate at a 3%/3mm gamma analysis. Conclusion: We have developed a new sequencing methodology capable of online plan adaptation. In this work, we extended the pipeline to support Pareto-optimal input and clinically validated that it can accurately achieve these ideal distributions, while its flexible design enables inter- and intrafraction plan adaptation. This research is financially supported by Elekta AB, Stockholm, Sweden.

60 APPLIED LIFE SCIENCES↗

DOE OSTI2023

Iterative Stability Enforcement in Adaptive Antoulas–Anderson Algorithms for ${\boldsymbol{\mathcal{H}_2}}$ Model Reduction

This paper presents an extension of the Adaptive-Antoulas-Anderson (AAA) algorithm for rational modelling. Specifically, our new stable multi-input multi-output AAA (smiAAA) algorithm builds rational approximations of multi-input signals with a common set of stable poles. A new methodology is presented for iteratively enforcing stability constraints on the poles. We demonstrate the strengths of this approach compared to the stability enforcement in the FastAAA algorithm. Results using the smiAAA algorithm are compared with the commonly used Vector Fitting algorithm and the more recently published RKFIT algorithm. Vector Fitting and RKFIT both require the user to input the number of poles to use in the approximations. If the final approximation is not accurate enough, the user must re-start Vector Fitting or RKFIT with a larger number of poles and/or a new starting location for the poles. In contrast, the smiAAA algorithm is designed to allow the user to simply input the desired accuracy of the approximations, and the necessary number of poles is detected automatically. This permits users to produce approximations of a desired accuracy with no knowledge about the underlying order of the system being approximated, preventing the algorithm from ever needing to be rerun. An additional feature for preventing extraneous poles from being returned by AAA is also discussed. The cause of these extraneous poles is efficiently detected and removed by our presented methodology. In conclusion, the examples presented demonstrate that smiAAA can efficiently produce approximations of similar or better accuracy than Vector Fitting and RKFIT while requiring less input from the user.

97 MATHEMATICS AND COMPUTING↗

DOE OSTI2024

ThunderBoltz API

The ThunderBoltz application programming interface (API) code is written in Python and is comprised of a set of tools to facilitate compilation of the ThunderBoltz code, as well as fast assembly and formatting of input files for the ThunderBoltz code, post-processing tools of ThunderBoltz results, plotting tools, and runs/schedules the ThunderBoltz code executable for calculations. The ThunderBoltz API code is utilized for importing and manipulating input cross section sets, input conditions, and any other simulation settings made available within the ThunderBoltz input deck via user-defined settings or via automatic generation. The API comes with a set of plotting capabilities of input cross sections, results from ThunderBoltz, and post-processed results carried out with the API.

Park, Ryan↗

DOE OSTI2025

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↗