Bayesian Sensitivity Analysis of a Cardiac Cell Model Using a Gaussian Process Emulator

Abstract. Sensitivity analysis of atmospheric models is necessary to identify the processes that lead to uncertainty in model predictions, to help understand model diversity, and to prioritise research. Assessing the effect of parameter uncertainty in complex models is challenging and often limited by CPU constraints. Here we present a cost-effective application of variance-based sensitivity analysis to quantify the sensitivity of a 3-D global aerosol model to uncertain parameters. A Gaussian process emulator is used to estimate the model output across multi-dimensional parameter space using information from a small number of model runs at points chosen using a Latin hypercube space-filling design. Gaussian process emulation is a Bayesian approach that uses information from the model runs along with some prior assumptions about the model behaviour to predict model output everywhere in the uncertainty space. We use the Gaussian process emulator to calculate the percentage of expected output variance explained by uncertainty in global aerosol model parameters and their interactions. To demonstrate the technique, we show examples of cloud condensation nuclei (CCN) sensitivity to 8 model parameters in polluted and remote marine environments as a function of altitude. In the polluted environment 95 % of the variance of CCN concentration is described by uncertainty in the 8 parameters (excluding their interaction effects) and is dominated by the uncertainty in the sulphur emissions, which explains 80 % of the variance. However, in the remote region parameter interaction effects become important, accounting for up to 40 % of the total variance. Some parameters are shown to have a negligible individual effect but a substantial interaction effect. Such sensitivities would not be detected in the commonly used single parameter perturbation experiments, which would therefore underpredict total uncertainty. Gaussian process emulation is shown to be an efficient and useful technique for quantifying parameter sensitivity in complex global atmospheric model.

Download Full-text

Emulation of a complex global aerosol model to quantify sensitivity to uncertain parameters

Atmospheric Chemistry and Physics ◽

10.5194/acp-11-12253-2011 ◽

2011 ◽

Vol 11 (23) ◽

pp. 12253-12273 ◽

Cited By ~ 87

Author(s):

L. A. Lee ◽

K. S. Carslaw ◽

K. J. Pringle ◽

G. W. Mann ◽

D. V. Spracklen

Keyword(s):

Sensitivity Analysis ◽

Gaussian Process ◽

Interaction Effects ◽

Model Parameters ◽

Uncertain Parameters ◽

Model Output ◽

Atmospheric Models ◽

Aerosol Model ◽

Gaussian Process Emulation ◽

Gaussian Process Emulator

Abstract. Sensitivity analysis of atmospheric models is necessary to identify the processes that lead to uncertainty in model predictions, to help understand model diversity through comparison of driving processes, and to prioritise research. Assessing the effect of parameter uncertainty in complex models is challenging and often limited by CPU constraints. Here we present a cost-effective application of variance-based sensitivity analysis to quantify the sensitivity of a 3-D global aerosol model to uncertain parameters. A Gaussian process emulator is used to estimate the model output across multi-dimensional parameter space, using information from a small number of model runs at points chosen using a Latin hypercube space-filling design. Gaussian process emulation is a Bayesian approach that uses information from the model runs along with some prior assumptions about the model behaviour to predict model output everywhere in the uncertainty space. We use the Gaussian process emulator to calculate the percentage of expected output variance explained by uncertainty in global aerosol model parameters and their interactions. To demonstrate the technique, we show examples of cloud condensation nuclei (CCN) sensitivity to 8 model parameters in polluted and remote marine environments as a function of altitude. In the polluted environment 95 % of the variance of CCN concentration is described by uncertainty in the 8 parameters (excluding their interaction effects) and is dominated by the uncertainty in the sulphur emissions, which explains 80 % of the variance. However, in the remote region parameter interaction effects become important, accounting for up to 40 % of the total variance. Some parameters are shown to have a negligible individual effect but a substantial interaction effect. Such sensitivities would not be detected in the commonly used single parameter perturbation experiments, which would therefore underpredict total uncertainty. Gaussian process emulation is shown to be an efficient and useful technique for quantifying parameter sensitivity in complex global atmospheric models.

Download Full-text

Advanced methods for uncertainty assessment and global sensitivity analysis of a Eulerian atmospheric chemistry transport model

10.5194/acp-2018-690 ◽

2018 ◽

Author(s):

Ksenia Aleksankina ◽

Stefan Reis ◽

Massimo Vieno ◽

Mathew R. Heal

Keyword(s):

Sensitivity Analysis ◽

Gaussian Process ◽

Atmospheric Chemistry ◽

Sampling Design ◽

Latin Hypercube Sampling ◽

Sensitivity Analyses ◽

Global Sensitivity ◽

Model Predictions ◽

Averaged Model ◽

Gaussian Process Emulator

Abstract. Atmospheric chemistry transport models (ACTMs) are extensively used to provide scientific support for the development of policies to mitigate against the detrimental effects of air pollution on human health and ecosystems. Therefore, it is essential to quantitatively assess the level of model uncertainty and to identify the model input parameters that contribute the most to the uncertainty. For complex process-based models, such as ACTMs, uncertainty and global sensitivity analyses are still challenging and are often limited by computational constraints due to the requirement of a large number of model runs. In this work, we demonstrate an emulator-based approach to uncertainty quantification and variance-based sensitivity analysis for the EMEP4UK model (regional application of the European Monitoring and Evaluation Programme Meteorological Synthesizing Centre-West). A separate Gaussian process emulator was used to estimate model predictions at unsampled points in the space of the uncertain model inputs for every modelled grid cell. The training points for the emulator were chosen using an optimised Latin hypercube sampling design. The uncertainties in surface concentrations of O3, NO2, and PM2.5 were propagated from the uncertainties in the anthropogenic emissions of NOx, SO2, NH3, VOC, and primary PM2.5 reported by the UK National Atmospheric Emissions Inventory. The results of the EMEP4UK uncertainty analysis for the annually averaged model predictions indicate that modelled surface concentrations of O3, NO2, and PM2.5 have the highest level of uncertainty in the grid cells comprising urban areas (up to ±7 %, ±9 %, and ±9 % respectively). The uncertainty in the surface concentrations of O3 and NO2 were dominated by uncertainties in NOx emissions combined from non-dominant sectors (i.e. all sectors excluding energy production and road transport) and shipping emissions. Additionally, uncertainty in OO3 was driven by uncertainty VOC emissions combined from sectors excluding solvent use. Uncertainties in the modelled PM2.5 concentrations were mainly driven by uncertainties in primary PM2.5 emissions and NH3 emissions from the agricultural sector. Uncertainty and sensitivity analyses were also performed for five selected grid sells for monthly averaged model predictions to illustrate the seasonal change in the magnitude of uncertainty and change in the contribution of different model inputs to the overall uncertainty. Our study demonstrates the viability of a Gaussian process emulator-based approach for uncertainty and global sensitivity analyses, which can be applied to other ACTMs. Conducting these analyses helps to increase the confidence in model predictions. Additionally, the emulators created for these analyses can be used to predict the ACTM response for any other combination of perturbed input emissions within the ranges set for the original Latin hypercube sampling design without the need to re-run the ACTM, thus allowing fast exploratory assessments at significantly reduced computational costs.

Download Full-text

Advanced methods for uncertainty assessment and global sensitivity analysis of an Eulerian atmospheric chemistry transport model

Atmospheric Chemistry and Physics ◽

10.5194/acp-19-2881-2019 ◽

2019 ◽

Vol 19 (5) ◽

pp. 2881-2898 ◽

Cited By ~ 7

Author(s):

Ksenia Aleksankina ◽

Stefan Reis ◽

Massimo Vieno ◽

Mathew R. Heal

Keyword(s):

Sensitivity Analysis ◽

Gaussian Process ◽

Atmospheric Chemistry ◽

Sampling Design ◽

Latin Hypercube Sampling ◽

Sensitivity Analyses ◽

Global Sensitivity ◽

Model Predictions ◽

Averaged Model ◽

Gaussian Process Emulator

Abstract. Atmospheric chemistry transport models (ACTMs) are extensively used to provide scientific support for the development of policies to mitigate the detrimental effects of air pollution on human health and ecosystems. Therefore, it is essential to quantitatively assess the level of model uncertainty and to identify the model input parameters that contribute the most to the uncertainty. For complex process-based models, such as ACTMs, uncertainty and global sensitivity analyses are still challenging and are often limited by computational constraints due to the requirement of a large number of model runs. In this work, we demonstrate an emulator-based approach to uncertainty quantification and variance-based sensitivity analysis for the EMEP4UK model (regional application of the European Monitoring and Evaluation Programme Meteorological Synthesizing Centre-West). A separate Gaussian process emulator was used to estimate model predictions at unsampled points in the space of the uncertain model inputs for every modelled grid cell. The training points for the emulator were chosen using an optimised Latin hypercube sampling design. The uncertainties in surface concentrations of O3, NO2, and PM2.5 were propagated from the uncertainties in the anthropogenic emissions of NOx, SO2, NH3, VOC, and primary PM2.5 reported by the UK National Atmospheric Emissions Inventory. The results of the EMEP4UK uncertainty analysis for the annually averaged model predictions indicate that modelled surface concentrations of O3, NO2, and PM2.5 have the highest level of uncertainty in the grid cells comprising urban areas (up to ±7 %, ±9 %, and ±9 %, respectively). The uncertainty in the surface concentrations of O3 and NO2 were dominated by uncertainties in NOx emissions combined from non-dominant sectors (i.e. all sectors excluding energy production and road transport) and shipping emissions. Additionally, uncertainty in O3 was driven by uncertainty in VOC emissions combined from sectors excluding solvent use. Uncertainties in the modelled PM2.5 concentrations were mainly driven by uncertainties in primary PM2.5 emissions and NH3 emissions from the agricultural sector. Uncertainty and sensitivity analyses were also performed for five selected grid cells for monthly averaged model predictions to illustrate the seasonal change in the magnitude of uncertainty and change in the contribution of different model inputs to the overall uncertainty. Our study demonstrates the viability of a Gaussian process emulator-based approach for uncertainty and global sensitivity analyses, which can be applied to other ACTMs. Conducting these analyses helps to increase the confidence in model predictions. Additionally, the emulators created for these analyses can be used to predict the ACTM response for any other combination of perturbed input emissions within the ranges set for the original Latin hypercube sampling design without the need to rerun the ACTM, thus allowing for fast exploratory assessments at significantly reduced computational costs.

Download Full-text