Search NASA⌕ Search

SEARCH · Search NASA

Results for “conditional generative models”

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 127 records · Page 7
All sourcesNASA NTRSDOE OSTI
DOE OSTI2022

Optimization control technology for building energy conservation

A simulation processor generates and stores a simulation model based on conditions associated with a physical structure, such as a building. A neural network processor implements a neural network, having an input layer coupled to receive sensor data from the structure and having an output layer coupled to supply control signals to the at least one electrically operable environmental control device. The neural network is trained using the simulation model. A particle swarm optimization processor programmed to receive the simulation results and perform particle swarm optimization, ascertains optimal parameters for controlling the at least one electrically operable environmental control device and supplies these optimal parameters to the neural network processor. The neural network processor uses the optimal parameters supplied by the particle swarm optimization processor to further train the neural network.

32 ENERGY CONSERVATION, CONSUMPTION, AND UTILIZATI↗
DOE OSTI2022

Analysis of contact conditions and microstructure evolution in shear assisted processing and extrusion using smoothed particle hydrodynamics method

Shear assisted processing and extrusion (ShAPE) is a solid-phase processing technique that adds an additional shear force as compared with a conventional extrusion approach. Recently, ShAPE has demonstrated the capability of extruding high-performance aluminum alloy 7075 (AA7075) tubes at speeds up to 12.2 m/min without surface tearing. However, the relationship among the ShAPE processing parameters, thermomechanical conditions, contact conditions, heat generation, and microstructure evolution remains primarily empirical because an insightful understanding of the associated physics is still lacking. To help elucidate these relationships, this work proposes a thermomechanical meshfree model for the first time for ShAPE processing of AA7075 using the smoothed particle hydrodynamics (SPH) method. The meshfree model is first validated thoroughly by experimental data in terms of material flow, die face temperature, and extrusion force with various processing parameters. The validated model is then used to analyze the steady-state contact conditions and heat generation rates during ShAPE processing. Distributions of the average grain size of AA7075 being extruded are calculated using the SPH model output. The meshfree model results reveal that extrusions conducted at lower temperatures and higher strain rates yield more refined grains and possibly higher material strength, which is also consistent with the experimental observations.

36 MATERIALS SCIENCE↗
DOE OSTI2021

Model, data, and code for paper "Modeling of streamflow in a 30-kilometer-long reach spanning 5 years using OpenFOAM 5.x"

The data package includes data, model, and code that support the analyses and conclusions in the paper titled “modeling of streamflow in a 30-kilometer-long reach spanning 5 years using OpenFOAM 5.x”. The primary goal of this paper is to demonstrate that key streamflow properties such as water depth, flow velocity, and dynamic pressure in a natural river at 30-kilometer scale over 5 years can be reliably and efficiently modeled using the computational framework presented in this paper. To support the paper, various data types from remote sensing, field observations, and computational models are used. Specific details are described as follows. Firstly, the river bathymetry data was obtained from a Light Detection and Ranging (LiDAR) survey. This data is then converted to a triangulated surface format, STL, for mesh generation in OpenFOAM. The STL data can be found in Model_Setups/BaseCase_2013To2015/constant/triSurface. The OpenFOAM mesh generated using this STL file can be found in constant/polyMesh. Other model setups, boundary and initial conditions can be found in /system and /0.org under folder BaseCase_2013To2015. A similar data structure can also be found in BaseCase_2018To2019 for the simulations during 2018 and 2019. Secondly, the OpenFOAM simulations need the upstream discharge and water depth information at the upstream boundary to drive the model. These data are generated from a one-dimensional hydraulic model and the data can be found under the folder Model_Setups /1D model Mass1 data. The mass1_65.csv and mass1_191.csv files include the results of the 1D model at the model inlet and outlet, respectively. The Matlab source code Mass1ToOFBC20182019.m is used to convert these data into OpenFOAM boundary condition setups.With the above OpenFOAM model, it can generate data for water surface elevation, flow velocity, and dynamic pressure. In this paper, the water surface elevation was measured at 7 locations during different periods between 2011 and 2019. The exact survey locations (see Fig1_SurveyLocations.txt) can be found in folder Fig_1. The variation of water stage over time at the 7 locations can be found in folder /Observation_WSE. The data type include .txt, .csv, .xlsx, and .mat. The .mat data can be loaded by Matlab.We also measured the flow velocities at 12 cross-sections along the river. At each cross-section, we recorded the x, y locations, depth, three velocity components u,v,w. These data are saved to a Matlab format which can be found under folder /Observation_Velocity and /Fig_1. The relative locations of velocity survey locations to the river bathymetry can be found in Figure 1c.The water stage data at the 7 locations from OpenFOAM, 1D, and 2D hydraulic models are also provided to evaluate the long-term performance of 3D models vs 1D/2D models. The water stage data for the 7 locations from OpenFOAM have been saved to .mat format and can be found in /OpenFOAM_WSE. The water stage data from the 1D model are saved in .csv format and can be found in /Mass1_WSE. The water stage from the 2D model is saved as .mat format and can be found in / Mass2_WSEIn addition, the OpenFOAM model outputs the information of hydrostatic and hydrodynamic pressure. They are saved as .mat format under folder /Fig_11/2013_1. As the files are too large, we only uploaded the data for January 2013. The area of different ratio of dynamic pressure to static pressure for all simulation range, i.e., 2013-2015, are saved to .mat format. They can be found in /Fig_11/PA. Further, the data of wall clock time versus the solution time of the OpenFOAM modeling are also saved to .mat format under folder /Fig_13/LogsMat. In summary, the data package contains seven data types, including .txt, .csv, .xlsx, .dat, .stl, .m, and .mat. The former 4 types can be directly open using a text editor or Microsoft Office. The .mat format needs to be read by Matlab. The Matlab source code .m files need to be run with Matlab. The OpenFOAM setups can be visualized in ParaView. The .stl file can be opened in ParaView or Blender. The data in subfolders Fig_1 to Fig_10 and Fig_12 are copied from the aforementioned data folders to generate specific figures for the paper. A readME.txt file is included in each subfolder to further describe how the data in each folder are generated and used to support the paper.Please use the data package's DOI to cite the data package. Please contact yunxiang.chen@pnnl.gov if you need more data related to the paper.

54 ENVIRONMENTAL SCIENCES↗
DOE OSTI2023

Deep generative models for vehicle speed trajectories

Generating realistic vehicle speed trajectories is a crucial component in evaluating vehicle fuel economy and in predictive control of self-driving cars. Traditional generative models rely on Markov chain methods and can produce accurate synthetic trajectories but are subject to the curse of dimensionality. They do not allow to include conditional input variables into the generation process. In this paper, we show how extensions to deep generative models allow accurate and scalable generation. Proposed architectures involve recurrent and feed-forward layers and are trained using adversarial techniques. Our models are shown to perform well on generating vehicle trajectories using a model trained on GPS data from Chicago metropolitan area.

33 ADVANCED PROPULSION SYSTEMS↗
DOE OSTI2022

Hybrid data-driven and model-informed online tool wear detection in milling machines

Precision machining tool wear is responsible for low product throughput and quality. Monitoring the tool wear online is vital to prevent degradation in machining quality. However, direct real-time tool wear measurement is not practical. This paper presents residual-based anomaly detection models, combining a hybrid model comprised of a physics-based model and a data-driven model (a decision tree or a neural network) to predict signals of interest (e.g., power or forces) under nominal conditions, followed by Page’s cumulative sum test for detecting tool wear on-line using the computer numerical control machine measurements. The most informative features are ranked using dynamic programming and its approximation variants from real-time measurements and machine settings, such as the width of cut, depth of cut, feed rate and spindle speed, that serve as inputs to the predictive models. The baseline nominal model is incrementally updated with experimental data via a gradient boosted adaptation model to generate the residuals that account for discrepancies between the actual machine data under normal conditions and the baseline nominal model predictions. The hybrid model is validated against 20 Mazak milling machine experimental tests and one Haas run-to-failure experiment. The proposed anomaly detector is applied to synthetic data from simulations of the physics-based model at different operating conditions, measurement noise levels, and tool wear levels, and the methods were able to achieve an overall 92% accuracy in data with 1% noise. The anomaly detection methods based on hybrid model reduced the false alarms of either the data-driven or physical-based models alone, and are found to be capable of good online detection of tool wear.

42 ENGINEERING↗
DOE OSTI2023

Modeling of Vertical Motor-driven Pump for Simulation of a Fault Signature \\ for Condition Monitoring

As part of the ongoing effort to transition from preventive maintenance strategies to condition-based maintenance strategies in nuclear power plants, there is significant reliance on using machine learning techniques. To develop a robust machine learning model that can diagnose all the fault modes of a vertical motor-driven pump, data capturing the unique signature of each fault mode is required. In practice, it is difficult to collect or capture data that captures all the fault modes from a single plant site. So to address this situation, a computational model of a vertical motor-driven pump is developed using the multipurpose finite element software COMSOL Multiphysics. The developed model is used to generate simulated data under normal operation and is compared with the vibration data collected using vibration sensors. Once the simulation model is verified under normal operating condition, simulated data for the fault mode for which minimal or no evidence is available in historical plant process data is developed. This simulated data is used to develop fault signatures to achieve robust predictive models. This paper presents modeling details and verification of the model that can used to generate data for fault modes that are not available at a plant site for condition monitoring purpose.

46 - INSTRUMENTATION RELATED TO NUCLEAR SCIENCE AN↗
DOE OSTI2022

Automated Control for Nuclear Thermal Propulsion Start-Up using MOOSE-based Applications

This report presents a Griffin/Bison/RELAP-7 numerical model of a prototypical NTP system that features fuel assemblies arranged in rings, and which was designed to simulate rapid startup transients. The physics modeled include full-core neutronics, assembly-wise heat conduction, and conjugate heat transfer, with the balance of plant mainly imposed through boundary conditions. In addition, various forms of automated reactivity control were deployed by using the MOOSE to autonomously drive the model and simulate the reactor transitioning from assumed initial conditions to nominal power in a fraction of a minute. To generate the cross-sections of the neutronics model, and in an effort to simultaneously account for the tremendous axial temperature gradients in the reactor and to limit the number of state points required for cross-section generation, the average component temperatures and hydrogen densities in the cooling channels were correlated to the average fuel and moderator temperatures, and fixed axial profiles were derived for nominal conditions and then used during the transient. With this approximation, a tractable cross-section library tabulated with fuel/moderator temperatures and CD angles was generated using Serpent. The full-core SPH correction procedure and the CD decusping technology in Griffin, respectively, ensure preservation of the multiplication factor and reaction rates at state points, along with a reasonably accurate reactivity worth between tabulated CD angles, despite using a coarse mesh. Feedback from other physics was calculated by modeling one representative fuel assembly per ring, along with the corresponding fuel and moderator cooling channels. To limit power overshoots during startup, another layer of multiphysics coupling was added to the model in order to automatically control the drums. Two different technologies presented herein showed outstanding performance in this regard: (1) a novel hybrid PID controller based on both power and reactivity signals, and (2) a PGC that relies on kinetics parameters and reactivity coefficients to predict future behavior and adjust the desired signal accordingly. A challenging benchmark was devised, featuring a power demand curve that exponentially increases by a factor of 500 within 30 seconds, then levels out after that. Both control approaches create a simulated power curve that closely follows the power demand curve and limits power overshoots to 1% or less. While the former approach requires more tuning of the internal parameters, the latter requires additional knowledge of the reactivity feedback coefficients and rates of change of the corresponding variables, including fuel and moderator temperature, which could be difficult to dynamically measure for a real NTP system. Fortunately, some inaccuracy in these quantities will not drastically degrade the PGC performance. Subsequently, a more realistic startup sequence was considered, in which the mass flow rate and outlet pressures are ramped up to model bootstrap and thrust build-up phases prior to reaching steady-state conditions, demonstrating the ability of the hybrid PID and PGCs to handle such transients, with both types of controllers exhibiting very similar behavior. Nevertheless, a significant chamber temperature overshoot was observed, caused by the demanded power signal and assumed mass flow rate. This issue could be mitigated by deploying a reactor controller that follows the chamber temperature signal and actuates both the control valves and drums (rather than using a power signal based solely on the drums to control reactivity). Enhancement of the hydrogen fluid properties available in MOOSE, as well as a better understanding of prototypical initial conditions, are also needed to further enhance this startup model. Finally, a study was performed to model decay heat post-shutdown, and to prepare for extending this model to predict shutdown behavior and post-shutdown pulsed cooling requirements.

33 ADVANCED PROPULSION SYSTEMS↗
DOE OSTI2019

Modified Paschen curves for pulsed breakdown

The influence of the voltage rise time on the Paschen curves is analyzed using kinetic and fluid models. The Paschen curves, which describe the dependence of the breakdown voltage, U br , on the product pd of gas pressure p and the characteristic length d between the electrodes, have minima that correspond to optimal conditions for plasma generation. It is shown that both kinetic and fluid models predict the experimentally observed shift of the curves toward higher voltages and the shift of the Paschen minima toward higher values of pd with the decreasing voltage rise time τ. On the right branch of the Paschen curves, the agreement between both models is obtained for all τ. We confirm that the minimum of the Paschen curves corresponds to conditions of electron runaway. Finally on the left branch, the electron velocity distribution function is nonlocal in space and also nonlocal in time.

71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSIC↗
DOE OSTI2015

Finite element model for piles in liquefiable ground

This paper develops a three dimensional finite element modelling method for piles in liquefiable ground and applies it to the analysis of seismic pile responses. A unified plasticity model for large post-liquefaction shear deformation of sand provides the basis for the effective and efficient modelling of liquefiable ground. Special attention is dedicated towards the modelling of piles and soil-pile interface to accurately reflect the behaviour of piles. A staged modelling procedure is adopted to appropriately generate the initial conditions for the soil and piles and achieve hydrostatic pore pressure prior to seismic loading. Three centrifuge shaking table tests on single piles, both with and without pile cap and superstructure, in level and inclined liquefiable ground are conducted and simulated in validation and application of the proposed method. Further studies to investigate the effects of pile cap, lateral spreading, and non-liquefiable surface layer are undertaken numerically using the validated method. The results show these aforementioned factors to be influential in the dynamic and residual response of piles in liquefiable ground.

Centrifuge test↗
DOE OSTI2015

A boundary condition to the Khokhlov-Zabolotskaya equation for modeling strongly focused nonlinear ultrasound fields

An equivalent source model was proposed as a boundary condition to the nonlinear parabolic Khokhlov-Zabolotskaya (KZ) equation to simulate high intensity focused ultrasound (HIFU) fields generated by medical ultrasound transducers with the shape of a spherical shell. The boundary condition was set in the initial plane; the aperture, the focal distance, and the initial pressure of the source were chosen based on the best match of the axial pressure amplitude and phase distributions in the Rayleigh integral analytic solution for a spherical transducer and the linear parabolic approximation solution for the equivalent source. Analytic expressions for the equivalent source parameters were derived. It was shown that the proposed approach allowed us to transfer the boundary condition from the spherical surface to the plane and to achieve a very good match between the linear field solutions of the parabolic and full diffraction models even for highly focused sources with F-number less than unity. The proposed method can be further used to expand the capabilities of the KZ nonlinear parabolic equation for efficient modeling of HIFU fields generated by strongly focused sources.

42 ENGINEERING↗
DOE OSTI2019

Impact of climate change on European winter and summer flood losses

Climate change is expected to alter European floods and associated economic losses in various ways. Here we investigate the impact of precipitation change on European average winter and summer financial losses due to flooding under a 1.5 °C warming scenario (reflecting a projected climate in the year 2115 according to RCP2.6)and for a counterfactual current-climate scenario where the climate has evolved without anthropogenic influence (reflecting a climate corresponding to pre-industrial conditions). Climate scenarios were generated with the Community Atmospheric Model (CAM)version 5. For each scenario, we derive a set of weights that when applied to the current climate's precipitation results in a climatology that approximates that of the scenario. We apply the weights to annual losses from a well-calibrated (to the current climate)flood loss model that spans 50,000 years and re-compute the average annual loss to assess the impact of precipitation changes induced by anthropogenic climate change. The method relies on a large stochastic set of physically based flood model simulations and allows quick assessment of potential loss changes due to change in precipitation based on two statistics, namely total precipitation, and total precipitation of very wet days (defined here as the total precipitation of days above the 95th percentile of daily precipitation). We compute the statistics with the raw CAM precipitation and bias-corrected precipitation. Our results show that for both raw and bias-corrected statistics i)average flood loss in Europe generally tend to increase in winter and decrease in summer for the future scenario, and consistent with that change we also show that ii)average flood losses have increased (decreased)for winter (summer)from pre-industrial conditions to the current day. Finally, the magnitude of the change varies among scenarios and statistics chosen.

54 ENVIRONMENTAL SCIENCES↗
DOE OSTI2026

Energy-Optimal Vehicle Longitudinal Motion Control via Pontryagin’s Minimum Principle and Ultra-Local Model

Longitudinal vehicle motion control is essential for enhancing performance and optimizing a vehicle’s energy usage. However, it remains a challenging task due to the nonlinear and uncertain nature of vehicle dynamics, along with varying driving conditions. This paper presents a novel ultra-local optimal control approach based on Pontryagin’s Minimum Principle (PMP) that circumvents the need for detailed system identification by employing an ultra-local model. The control objective is to minimize the total energy consumption under boundary conditions while ensuring smooth traction force generation. The proposed approach is evaluated using a high-fidelity vehicle model in three representative scenarios: (i) nominal driving, (ii) a change in tire road friction coefficient (TRFC) from 0.5 to 0.65 and road slope from 0% to 5% during the maneuver, with target velocity unchanged, and (iii) a change in target velocity from 20 m/s to 0 m/s during the maneuver, while maintaining nominal TRFC and slope conditions. The simulation results demonstrate that the proposed method delivers robust performance, effectively balancing consumption and tracking accuracy in all tested scenarios.

Waleed khan, Muhammad [The University of Texas at ↗
DOE OSTI2025

Modeling of Vertical Motor-driven Pump for Simulation of a Fault Signature \\ for Condition Monitoring

As part of the ongoing effort to transition from preventive maintenance strategies to condition-based maintenance strategies in nuclear power plants, there is significant reliance on using machine learning techniques. To develop a robust machine learning model that can diagnose all the fault modes of a vertical motor-driven pump, data capturing the unique signature of each fault mode is required. In practice, it is difficult to collect or capture data that captures all the fault modes from a single plant site. So to address this situation, a computational model of a vertical motor-driven pump is developed using the multipurpose finite element software COMSOL Multiphysics. The developed model is used to generate simulated data under normal operation and is compared with the vibration data collected using vibration sensors. Once the simulation model is verified under normal operating condition, simulated data for the fault mode for which minimal or no evidence is available in historical plant process data is developed. This simulated data is used to develop fault signatures to achieve robust predictive models. This paper presents modeling details and verification of the model that can used to generate data for fault modes that are not available at a plant site for condition monitoring purpose.

46 - INSTRUMENTATION RELATED TO NUCLEAR SCIENCE AN↗
DOE OSTI2024

Denoising diffusion probabilistic models for generative alloy design

Inverse material design is an extremely challenging optimization task made difficult by, in part, the highly nonlinear relationship linking performance with composition. Quantitative approaches have improved significantly owing to advances in high throughput experimentation and computational thermodynamics. However, existing physics-based tools are mostly forward models; input a chemistry and obtain a prediction. More recently the materials community has leveraged advances in the machine learning community to establish novel inverse design frameworks. Very recently denoising diffusion probabilistic models have been shown to be extremely powerful generators producing synthetic data of various modalities e.g. images, text, audio, tables, etc.. In this work a novel framework for alloy design and optimization is proposed leveraging these class of models. Five key generative tasks are demonstrated (1) unconditional generation (2) composition conditioned generation (3) property conditioned generation (4) multi-feedstock conditioned generation and (5) generative optimization. These methods were tested on three case studies: high entropy alloy design, superalloy binder jet additive manufacturing, and in-situ dual-feedstock wire-arc additive manufacturing. Results indicate that the established models are extremely flexible, expressive, and robust. The architecture’s flexibility and training procedure empower the model to learn complex intra-compositional and composition-property relationships. Furthermore, the probabilistic nature of these models makes them well suited for addressing solution non-uniqueness and tackling uncertainty quantification tasks. While the fidelity and quantity of the underlying training data is paramount, we envision that future alloy design frameworks will make extensive use of these kinds of machine learning models as “search” tools bolstering the utility of experimental and computational approaches.

36 MATERIALS SCIENCE↗
DOE OSTI2024

Spectroscopy-guided discovery of three-dimensional structures of disordered materials with diffusion models

Spectroscopy techniques such as x-ray absorption near edge structure (XANES) provide valuable insights into the atomic structures of materials, yet the inverse prediction of precise structures from spectroscopic data remains a formidable challenge. In this study, we introduce a framework that combines generative artificial intelligence models with XANES spectroscopy to predict three-dimensional atomic structures of disordered systems, using amorphous carbon (a-C) as a model system. In this work, we introduce a new framework based on the diffusion model, a recent generative machine learning method, to predict 3D structures of disordered materials from a target property. For demonstration, we apply the model to identify the atomic structures of a-C as a representative material system from the target XANES spectra. We show that conditional generation guided by XANES spectra reproduces key features of the target structures. Furthermore, we show that our model can steer the generative process to tailor atomic arrangements for a specific XANES spectrum. Finally, our generative model exhibits a remarkable scale-agnostic property, thereby enabling generation of realistic, large-scale structures through learning from a small-scale dataset (i.e. with small unit cells). Our work represents a significant stride in bridging the gap between materials characterization and atomic structure determination; in addition, it can be leveraged for materials discovery in exploring various material properties as targeted.

36 MATERIALS SCIENCE↗
DOE OSTI2025

Automating and Evaluating Large Language Models for Accurate Text Summarization Under Zero-Shot Conditions

Automated text summarization (ATS) is crucial for collecting specialized, domain-specific information. Zero-shot learning (ZSL) allows large language models (LLMs) to respond to prompts on information not included in their training, playing a vital role in this process. This study evaluates LLMs' effectiveness in generating accurate summaries under ZSL conditions and explores using retrieval augmented generation (RAG) and prompt engineering to enhance factual accuracy and understanding. We combined LLMs with summarization modeling, prompt engineering, and RAG, evaluating the summaries using the METEOR metric and keyword frequencies through word clouds. Results indicate that LLMs are generally well-suited for ATS tasks, demonstrating an ability to handle specialized information under ZSL conditions with RAG. However, web scraping limitations hinder a single generalized retrieval mechanism. While LLMs show promise for ATS under ZSL conditions with RAG, challenges like goal misgeneralization and web scraping limitations need addressing. Future research should focus on solutions to these issues.

Priebe Mendes Rocha, Maria Eduarda [ORNL]↗
DOE OSTI2026

A bi-level advanced control framework for large-scale control of buildings with system-level impact

Increased electricity consumption combined with new forms of generation is testing the reliability of our grid infrastructure. This work describes a method to improve the reliability of the grid through large-scale advanced building control. This paper develops a bi-level distributed control framework to shift the load of 153 buildings to achieve a system-level objective of tracking a power reference signal. This bi-level control is based on the previously-developed ANPV-MPC, a predictive controller that uses a Bayesian neural network to generate an accurate control model and adapt to changing conditions over time. By shifting the building electricity demand to better match the available power, the grid system supplying the buildings is more reliable as evidenced by the analysis of node voltages across an IEEE 13-bus distribution system. The proposed bi-level control framework tracks the system-level power reference with enough accuracy to regulate node voltages across the IEEE 13-bus distribution system within ANSI limits of ±5%. Additionally, the adaptive nature of ANPV-MPC allows each building across the system to adapt to changing conditions, further amplifying the system-level reliability.

32 ENERGY CONSERVATION, CONSUMPTION, AND UTILIZATI↗
DOE OSTI2024

A comparison of model validation approaches for echo state networks using climate model replicates

As global temperatures continue to rise, climate mitigation strategies such as stratospheric aerosol injections (SAI) are increasingly discussed, but the downstream effects of these strategies are not well understood. As such, there is interest in developing statistical methods to quantify the evolution of climate variable relationships during the time period surrounding an SAI. Feature importance applied to echo state network (ESN) models has been proposed as a way to understand the effects of SAI using a data-driven model. This approach depends on the ESN fitting the data well. If not, the feature importance may place importance on features that are not representative of the underlying relationships. Typically, time series prediction models such as ESNs are assessed using out-of-sample performance metrics that divide the times series into separate training and testing sets. However, this model assessment approach is geared towards forecasting applications and not scenarios such as the motivating SAI example where the objective is using a data driven model to capture variable relationships. Here, in this paper, we demonstrate a novel use of climate model replicates to investigate the applicability of the commonly used repeated hold-out model assessment approach for the SAI application. Simulations of an SAI are generated using a simplified climate model, and different initialization conditions are used to provide independent training and testing sets containing the same SAI event. The climate model replicates enable out-of-sample measures of model performance, which are compared to the single time series hold-out validation approach. For our case study, it is found that the repeated hold-out sample performance is comparable, but conservative, to the replicate out-of-sample performance when the training set contains enough time after the aerosol injection.

54 ENVIRONMENTAL SCIENCES↗