Search NASA⌕ Search

SEARCH · Search NASA

Results for “Model Uncertainty”

Search indexed NASA NTRS and DOE OSTI research on propulsion, heat transfer, battery materials and energy systems. Follow report and document links to the original sources.

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

At least 1,081 records · Page 60
All sourcesNASA NTRSDOE OSTI
DOE OSTI2022

Sparsifying priors for Bayesian uncertainty quantification in model discovery

We propose a probabilistic model discovery method for identifying ordinary differential equations governing the dynamics of observed multivariate data. Our method is based on the sparse identification of nonlinear dynamics (SINDy) framework, where models are expressed as sparse linear combinations of pre-specified candidate functions. Promoting parsimony through sparsity leads to interpretable models that generalize to unknown data. Instead of targeting point estimates of the SINDy coefficients, we estimate these coefficients via sparse Bayesian inference. The resulting method, uncertainty quantification SINDy (UQ-SINDy), quantifies not only the uncertainty in the values of the SINDy coefficients due to observation errors and limited data, but also the probability of inclusion of each candidate function in the linear combination. UQ-SINDy promotes robustness against observation noise and limited data, interpretability (in terms of model selection and inclusion probabilities) and generalization capacity for out-of-sample forecast. Sparse inference for UQ-SINDy employs Markov chain Monte Carlo, and we explore two sparsifying priors: the spike and slab prior, and the regularized horseshoe prior. UQ-SINDy is shown to discover accurate models in the presence of noise and with orders-of-magnitude less data than current model discovery methods, thus providing a transformative method for real-world applications which have limited data.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2012

Quantifying uncertainty in LCA-modelling of waste management systems

Highlights: Black-Right-Pointing-Pointer Uncertainty in LCA-modelling of waste management is significant. Black-Right-Pointing-Pointer Model, scenario and parameter uncertainties contribute. Black-Right-Pointing-Pointer Sequential procedure for quantifying uncertainty is proposed. Black-Right-Pointing-Pointer Application of procedure is illustrated by a case-study. - Abstract: Uncertainty analysis in LCA studies has been subject to major progress over the last years. In the context of waste management, various methods have been implemented but a systematic method for uncertainty analysis of waste-LCA studies is lacking. The objective of this paper is (1) to present the sources of uncertainty specifically inherent to waste-LCA studies, (2) to select and apply several methods for uncertainty analysis and (3) to develop a general framework for quantitative uncertainty assessment of LCA of waste management systems. The suggested method is a sequence of four steps combining the selected methods: (Step 1) a sensitivity analysis evaluating the sensitivities of the results with respect to the input uncertainties, (Step 2) an uncertainty propagation providing appropriate tools for representing uncertainties and calculating the overall uncertainty of the model results, (Step 3) an uncertainty contribution analysis quantifying the contribution of each parameter uncertainty to the final uncertainty and (Step 4) as a new approach, a combined sensitivity analysis providing a visualisation of the shift in the ranking of different options due to variations of selected key parameters. This tiered approach optimises the resources available to LCA practitioners by only propagating the most influential uncertainties.

12 MANAGEMENT OF RADIOACTIVE AND NON-RADIOACTIVE W↗
DOE OSTI2021

Bayesian automated weighting of aggregated DFT, MD, and experimental data for candidate thermodynamic models of aluminum with uncertainty quantification

Atomic-scale modeling methods such as density functional theory (DFT) and molecular dynamics (MD) can predict the thermodynamic properties of materials at a lower cost than experimental measurements. However, their regular usage in thermodynamic model construction is hampered by the lack of quantitative agreement with experimental measurements and the lack of uncertainty estimates on the data. To make regular usage of this atomistic simulation data, it is important to assess whether the atomistic simulation datasets, by themselves or in combination with experimental measurements, result in the same physics-informed models best supported by experimental measurements alone. Here, models of aluminum thermodynamic properties are discussed using three data sources: atomistic calculations (DFT and MD), experiments, and a combination of atomistic calculations and experiments. The study shows that, after ensuring self-consistency in predicting key invariant points, both experimental measurements and atomistic calculations can significantly contribute to an optimal model.

36 MATERIALS SCIENCE↗
DOE OSTI2017

Bayesian calibration of groundwater models with input data uncertainty

Effective water resources management typically relies on numerical models to analyze groundwater flow and solute transport processes. Groundwater models are often subject to input data uncertainty, as some inputs (such as recharge and well pumping rates) are estimated and subject to uncertainty. Current practices of groundwater model calibration often overlook uncertainties in input data; this can lead to biased parameter estimates and compromised predictions. Through a synthetic case study of surface–ground water interaction under changing pumping conditions and land use, we investigate the impacts of uncertain pumping and recharge rates on model calibration and uncertainty analysis. We then present a Bayesian framework of model calibration to handle uncertain input of groundwater models. The framework implements a marginalizing step to account for input data uncertainty when evaluating likelihood. It was found that not accounting for input uncertainty may lead to biased, overconfident parameter estimates because parameters could be over–adjusted to compensate for possible input data errors. Parameter compensation can have deleterious impacts when the calibrated model is used to make forecast under a scenario that is different from calibration conditions. By marginalizing input data uncertainty, the Bayesian calibration approach effectively alleviates parameter compensation and gives more accurate predictions in the synthetic case study. The marginalizing Bayesian method also decomposes prediction uncertainty into uncertainties contributed by parameters, input data, and measurements. Furthermore, the results underscore the need to account for input uncertainty to better inform postmodeling decision making.

54 ENVIRONMENTAL SCIENCES↗
DOE OSTI2021

Earth System Model Improvement Pipeline via Uncertainty Attribution and Active Learning

Primary focal area: 2 (Predictive Modeling via AI): We develop methods to formally quantify uncertainties in Earth System models for the land-atmosphere coupled system. Science Challenge: Earth system models still have significant biases in historical predictions of the intensity and frequency of water cycling extremes (e.g., droughts and flood events), leading to low confidence in future projections. Uncertainties arise from incomplete understanding of land and atmospheric processes, and insufficient observational constraints on model parameters. Many observations, including those from key DOE investments such as ARM and AmeriFlux, are used to evaluate model performance but have not been used to formally quantify model uncertainty because of the expense of running ESM simulations. An efficient pipeline engaging cutting-edge machine learning (ML) and uncertainty quantification (UQ) methods is needed to improve the predictive understanding of water cycle extremes in the Earth system.

54 ENVIRONMENTAL SCIENCES↗
NASA NTRS2015

Random Predictor Models for Rigorous Uncertainty Quantification: Part 2

This and a companion paper propose techniques for constructing parametric mathematical models describing key features of the distribution of an output variable given input-output data. By contrast to standard models, which yield a single output value at each value of the input, Random Predictors Models (RPMs) yield a random variable at each value of the input. Optimization-based strategies for calculating RPMs having a polynomial dependency on the input and a linear dependency on the parameters are proposed. These formulations yield RPMs having various levels of fidelity in which the mean, the variance, and the range of the model's parameter, thus of the output, are prescribed. As such they encompass all RPMs conforming to these prescriptions. The RPMs are optimal in the sense that they yield the tightest predictions for which all (or, depending on the formulation, most) of the observations are less than a fixed number of standard deviations from the mean prediction. When the data satisfies mild stochastic assumptions, and the optimization problem(s) used to calculate the RPM is convex (or, when its solution coincides with the solution to an auxiliary convex problem), the model's reliability, which is the probability that a future observation would be within the predicted ranges, is bounded rigorously.

Crespo, Luis G.↗
NASA NTRS2015

Random Predictor Models for Rigorous Uncertainty Quantification: Part 1

This and a companion paper propose techniques for constructing parametric mathematical models describing key features of the distribution of an output variable given input-output data. By contrast to standard models, which yield a single output value at each value of the input, Random Predictors Models (RPMs) yield a random variable at each value of the input. Optimization-based strategies for calculating RPMs having a polynomial dependency on the input and a linear dependency on the parameters are proposed. These formulations yield RPMs having various levels of fidelity in which the mean and the variance of the model's parameters, thus of the predicted output, are prescribed. As such they encompass all RPMs conforming to these prescriptions. The RPMs are optimal in the sense that they yield the tightest predictions for which all (or, depending on the formulation, most) of the observations are less than a fixed number of standard deviations from the mean prediction. When the data satisfies mild stochastic assumptions, and the optimization problem(s) used to calculate the RPM is convex (or, when its solution coincides with the solution to an auxiliary convex problem), the model's reliability, which is the probability that a future observation would be within the predicted ranges, can be bounded tightly and rigorously.

Crespo, Luis G.↗
DOE OSTI2025

Simulating Atmospheric Processes in Earth System Models and Quantifying Uncertainties With Deep Learning Multi‐Member and Stochastic Parameterizations

Abstract Deep learning is a powerful tool to represent subgrid processes in climate models, but many application cases have so far used idealized settings and deterministic approaches. Here, we develop stochastic parameterizations with calibrated uncertainty quantification to learn subgrid convective and turbulent processes and surface radiative fluxes of a superparameterization embedded in an Earth System Model (ESM). We explore three methods to construct stochastic parameterizations: (a) a single Deep Neural Network (DNN) with Monte Carlo Dropout; (b) a multi‐member parameterization; and (c) a Variational Encoder Decoder with latent space perturbation. We show that the multi‐member parameterization improves the representation of convective processes, especially in the planetary boundary layer, compared to individual DNNs. The respective uncertainty quantification illustrates that methods (b) and (c) are advantageous compared to a dropout‐based DNN parameterization regarding the spread of convective processes. Hybrid simulations with our best‐performing multi‐member parameterizations remained challenging and crash within the first days. Therefore, we develop a pragmatic partial coupling strategy relying on the superparameterization for condensate emulation. Partial coupling reduces the computational efficiency of hybrid Earth‐like simulations but enables model stability over 5 months with our multi‐member parameterizations. However, our hybrid simulations exhibit biases in thermodynamic fields and differences in precipitation patterns. Despite this, the multi‐member parameterizations enable improvements in reproducing tropical extreme precipitation compared to a traditional convection parameterization. Despite these challenges, our results indicate the potential of a new generation of multi‐member machine learning parameterizations leveraging uncertainty quantification to improve the representation of stochasticity of subgrid effects.

Behrens, Gunnar [Deutsches Zentrum für Luft‐ und R↗
DOE OSTI2022

Kinetic model development and Bayesian uncertainty quantification for the complete reduction of Fe-based oxygen carriers with CH 4 , CO, and H 2 for chemical looping combustion

In this work, three kinetic models are developed and calibrated for the complete multi-step reduction of an Fe-based oxygen carrier (OC) particle with CH 4 , CO, and H 2 , using data from thermogravimetric analysis. The complete reduction rate profiles exhibit complex dynamics whose trajectory is significantly different depending on the reducing gas. A Bayesian model building and parameter estimation framework is applied for simultaneous parameter and model structure uncertainty quantification. The final models show excellent agreement between model predictions and calibration data, as well as new data not used for calibration (for the reduction of the OC with CH 4 ). Parameter uncertainty is quantified by determining joint posterior distribution, and model structure uncertainty is addressed by incorporating Gaussian process stochastic functions (represented by Bayesian smoothing splines) into the kinetic models. The final kinetic models with discrepancy functions are readily employable in equation-oriented simulation and optimization platforms.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗
DOE OSTI2014

A New Perspective for the Calibration of Computational Predictor Models.

This paper presents a framework for calibrating computational models using data from sev- eral and possibly dissimilar validation experiments. The offset between model predictions and observations, which might be caused by measurement noise, model-form uncertainty, and numerical error, drives the process by which uncertainty in the models parameters is characterized. The resulting description of uncertainty along with the computational model constitute a predictor model. Two types of predictor models are studied: Interval Predictor Models (IPMs) and Random Predictor Models (RPMs). IPMs use sets to characterize uncer- tainty, whereas RPMs use random vectors. The propagation of a set through a model makes the response an interval valued function of the state, whereas the propagation of a random vector yields a random process. Optimization-based strategies for calculating both types of predictor models are proposed. Whereas the formulations used to calculate IPMs target solutions leading to the interval value function of minimal spread containing all observations, those for RPMs seek to maximize the models' ability to reproduce the distribution of obser- vations. Regarding RPMs, we choose a structure for the random vector (i.e., the assignment of probability to points in the parameter space) solely dependent on the prediction error. As such, the probabilistic description of uncertainty is not a subjective assignment of belief, nor is it expected to asymptotically converge to a fixed value, but instead it is a description of the model's ability to reproduce the experimental data. This framework enables evaluating the spread and distribution of the predicted response of target applications depending on the same parameters beyond the validation domain (i.e., roll-up and extrapolation).

Crespo, Luis Guillermo↗
DOE OSTI2020

CO 2 plume evolution in a depleted natural gas reservoir: Modeling of conformance uncertainty reduction over time

Uncertainty in the long-term fate of CO 2 injected for geologic carbon sequestration (GCS) is a significant barrier to the adoption of GCS as a greenhouse-gas emission-mitigation for industry and regulatory agencies alike. Here, we present a modeling study that demonstrates that the uncertainty in forecasts of GCS site performance decreases over time as monitoring data are used to update operational models. We consider a case study of GCS in a depleted natural gas reservoir, with CO 2 injection occurring over 20 years, with a 50-year post-injection site care period. We constructed a detailed model to generate the actual model output, which is considered synthetic observation data. A series of simpler operational models based on limited data and assumptions about how an operator would model such a site are then run and compared against actual model output at specific monitoring points after one year, two years, etc. The operational model is updated and improved using the synthetic observation data from the actual model at the same time intervals. Model parameter values and model features needed to be updated over time to improve matches to the actual model. These kinds of model adjustments would be a normal part of reservoir engineering and site management at GCS sites. Uncertainty in two key measures related to site performance decreases with time: extent of the CO 2 plume up-dip migration, and radial extent of the pressure pulse. This conclusion should help allay the concerns of industry and regulators about uncertainty in long-term fate of CO 2 at GCS sites.

54 ENVIRONMENTAL SCIENCES↗
DOE OSTI2021

Gap-filling eddy covariance methane fluxes: Comparison of machine learning model predictions and uncertainties at FLUXNET-CH4 wetlands

Time series of methane fluxes measured by eddy-covariance require gap-filling to estimate annual emissions. Gap-filling methane fluxes is challenging because of high variability and complex responses to multiple drivers. To date, there is no widely established gap-filling standard for methane, with regards both to the best model algorithms and predictors. In this study, we address the need for standardization by synthesizing results of gap-filling methods applied at 17 wetland sites spanning boreal to tropical regions including all major wetlands classes and two rice paddies. We introduce new procedures for: 1) creating realistic artificial gap scenarios, 2) training and evaluating gap-filling models without overstating performance, and 3) predicting half-hourly methane fluxes and annual emissions with robust uncertainty estimates. We tested a conventional method (marginal distribution sampling) and four machine learning algorithms - penalized linear regression, artificial neural networks, random forests, and boosted decision trees - and four predictor sets, including temporal, meteorological, ecosystem carbon and energy flux, and soil predictors. We find that the conventional method can achieve similar median performance to the machine learning models but is worse than the best machine learning models and relatively insensitive to predictor choices. Of the machine learning models, decision tree algorithms performed the best in cross-validation experiments, even with a baseline predictor set, and artificial neural networks showed comparable performance when using all predictors. Soil temperature was frequently the most important predictor whilst water table depth was important at sites with substantial water table fluctuations, highlighting the value of data on soil conditions. Raw gap-filling uncertainties from the machine learning models were underestimated and we propose a method to calibrate uncertainties to observations. Finally, we gap-fill and provide summary evaluation metrics for all 81 sites in the FLUXNET-CH4 community dataset and publicly release the python code for model development, evaluation, and uncertainty estimation.

42 ENGINEERING↗
DOE OSTI2019

Carbon Dioxide Plume Evolution Following Injection into a Depleted Natural Gas Reservoir: Modeling of Conformance Uncertainty Reduction Over Time

The uncertainty in the long-term fate of CO 2 injected for geologic carbon sequestration (GCS) is a significant barrier to the adoption of GCS as a greenhouse gas emission mitigation approach for industry and regulatory agencies alike. Here we present a modeling study that demonstrates that the uncertainty in forecasts of GCS site performance decreases over time as monitoring data are used to inform and update operational models. The approach we take is to consider a case study consisting of a depleted natural gas reservoir that is used for GCS with CO 2 injection occurring over 20 years, with a 50-year post-injection site care (PISC) period. We constructed a detailed model of the system and ran this model out to 200 years to generate the actual site data. A series of simpler operational models based on limited data and assumptions about how an actual operator would model such a site are then run and compared against the actual model output at various specific monitoring points after one year, two years, etc. The operational model is then updated and improved using the observations (synthetic data from the actual model) at the same time intervals. We found that both model parameter values and model features needed to be added over time to improve matches to the actual system. These kinds of model adjustments are expected to be a normal part of reservoir engineering and site management at GCS sites. We found that the uncertainty in two key measures related to site performance at various locations decreases with time. This overall conclusion should help allay the concerns of industry and regulators about the uncertainty in GCS operations.

03 NATURAL GAS↗
DOE OSTI2015

Sparing Healthy Tissue and Increasing Tumor Dose Using Bayesian Modeling of Geometric Uncertainties for Planning Target Volume Personalization

Objective: To develop a mathematical tool that can update a patient's planning target volume (PTV) partway through a course of radiation therapy to more precisely target the tumor for the remainder of treatment and reduce dose to surrounding healthy tissue. Methods and Materials: Daily on-board imaging was used to collect large datasets of displacements for patients undergoing external beam radiation therapy for solid tumors. Bayesian statistical modeling of these geometric uncertainties was used to optimally trade off between displacement data collected from previously treated patients and the progressively accumulating data from a patient currently partway through treatment, to optimally predict future displacements for that patient. These predictions were used to update the PTV position and margin width for the remainder of treatment, such that the clinical target volume (CTV) was more precisely targeted. Results: Software simulation of dose to CTV and normal tissue for 2 real prostate displacement datasets consisting of 146 and 290 patients treated with a minimum of 30 fractions each showed that re-evaluating the PTV position and margin width after 8 treatment fractions reduced healthy tissue dose by 19% and 17%, respectively, while maintaining CTV dose. Conclusion: Incorporating patient-specific displacement patterns from early in a course of treatment allows PTV adaptation for the remainder of treatment. This substantially reduces the dose to healthy tissues and thus can reduce radiation therapy–induced toxicities, improving patient outcomes.

62 RADIOLOGY AND NUCLEAR MEDICINE↗