Calibrating a global atmospheric chemistry transport model using Gaussian process emulation and ground-level concentrations of ozone and carbon monoxide

Atmospheric chemistry transport models are important tools to investigate the local, regional and global controls on atmospheric composition and air quality. To ensure that these models represent the atmosphere adequately, it is important to compare their outputs with measurements. However, ground based measurements of atmospheric composition are typically sparsely distributed and representative of much smaller spatial scales than those resolved in models; thus, direct comparison incurs uncertainty. In this study, we investigate the feasibility of using observations of one or more atmospheric constituents to estimate parameters in chemistry transport models and to explore how these estimates and their uncertainties depend upon representation errors and the level of spatial coverage of the measurements. We apply Gaussian process emulation to explore the model parameter space and use monthly averaged ground-level concentrations of ozone (O3) and carbon monoxide (CO) from across Europe and the US. Using synthetic observations, we find that the estimates of parameters with greatest influence on O3 and CO are unbiased, and the associated parameter uncertainties are low even at low spatial coverage or with high representation error. Using reanalysis data, we find that estimates of the most influential parameter – corresponding to the dry deposition process – are closer to its expected value using both O3 and CO data than using O3 alone. This is remarkable because it shows that while CO is largely unaffected by dry deposition, the additional constraints it provides are valuable for achieving unbiased estimates of the dry deposition parameter. In summary, these findings identify the level of spatial representation error and coverage needed to achieve good parameter estimates and highlight the benefits of using multiple constraints to calibrate atmospheric chemistry transport models.

Global Optimization of Cultivar Trait Parameters in the Simulation of Sugarcane Phenology Using Gaussian Process Emulation

The global optimization of parameters in process-based crop models is often considered computationally expensive. Gaussian process (GP) emulation is a widely used method for reducing the computational burden of the optimization process. Total above-ground biomass and cane dry weight of three Thai sugarcane cultivars (KK3, LK92-11 and 02-2-058) collected under rainfed and irrigated conditions were used to optimize cultivar-specific parameters in the Agricultural Production Systems sIMulator (APSIM)-Sugarcane crop model through a GP emulation. GP emulators were trained and validated to approximate APSIM-Sugarcane model and then used for optimizing the cultivar-specific parameters through the differential evolution algorithm. Resulting optimized parameters allowed to obtain simulations that quite well approximated the observed biomass and CDW (validation results between simulated and observed yields: R2 0.93–0.98; normalized root mean squared error: 5–22%; Willmott’s agreement index: 0.87–0.99). The best parametrization was obtained under the lowest water stressed conditions. Based on these results, we suggest that GP emulation can be efficiently implemented for the parameterization of computationally expensive simulators.

Gaussian Process emulation of multi-model ice sheet and glacier projections

<p><strong>The land ice contribution to global mean sea level rise has not yet been predicted with ice sheet and glacier models for the latest set of socio-economic scenarios (SSPs), nor with coordinated exploration of uncertainties arising from the various computer models involved. Two recent international projects (ISMIP6 and GlacierMIP) generated a large suite of projections using multiple models, but mostly used previous generation scenarios and climate models, and could not fully explore known uncertainties. </strong></p><p><strong>Here we estimate probability distributions for these projections for the SSPs using Gaussian Process emulation of the ice sheet and glacier model ensembles. We model the sea level contribution as a function of global mean surface air temperature forcing and (for the ice sheets) model parameters, with the 'nugget' allowing for multi-model structural uncertainty. Approximate independence of ice sheet and glacier models is assumed, because a given model responds very differently under different setups (such as initialisation). </strong></p><p><strong>We find that limiting global warming to 1.5</strong>°<strong>C </strong><strong>would halve the land ice contribution to 21<sup>st</sup> century </strong><strong>sea level rise</strong><strong>, relative to current emissions pledges: t</strong><strong>he median decreases from 25 to 13 cm sea level equivalent (SLE) by 2100. However, the Antarctic contribution does not show a clear response to emissions scenario, due to competing processes of increasing ice loss and snowfall accumulation in a warming climate. </strong></p><p><strong>However, under risk-averse (pessimistic) assumptions for climate and Antarctic ice sheet model selection and ice sheet model parameter values, Antarctic ice loss could be five times higher, increasing the median land ice contribution to 42 cm SLE under current policies and pledges, with the 95<sup>th</sup> percentile exceeding half a metre even under 1.5</strong>°<strong>C warming. </strong></p><p><strong>Gaussian Process emulation can therefore be a powerful tool for estimating probability density functions from multi-model ensembles and testing the sensitivity of the results to assumptions.</strong></p>

Calibrating a global atmospheric chemistry transport model using Gaussian process emulation and ground-level concentrations of ozone and carbon monoxide

Atmospheric chemistry transport models are important tools to investigate the local, regional and global controls on atmospheric composition and air quality. To ensure that these models represent the atmosphere adequately it is important to compare their outputs with measurements. However, ground based measurements of atmospheric composition are typically sparsely distributed and representative of much smaller spatial scales than those resolved in models, and thus direct comparison incurs uncertainty. In this study, we investigate the feasibility of using observations of one or more atmospheric constituents to estimate parameters in chemistry transport models and to explore how these estimates and their uncertainties depend upon representation errors and the level of spatial coverage of the measurements. We apply Gaussian process emulation to explore the model parameter space and use monthly averaged ground-level concentrations of ozone (O3) and carbon monoxide (CO) from across Europe and the US. Using synthetic observations we find that the estimates of parameters with greatest influence on O3 and CO are unbiased, and the associated parameter uncertainties are low even at low spatial coverage or with high representation error. Using reanalysis data, we find that estimates of the most influential parameter – corresponding to the dry deposition process – are closer to its expected value using both O3 and CO data than using O3 alone. This is remarkable because it shows that while CO is largely unaffected by dry deposition, the additional constraints it provides are valuable for achieving unbiased estimates of the dry deposition parameter. In summary, these findings identify the level of spatial representation error and coverage needed to achieve good parameter estimates and highlight the benefits of using multiple constraints to calibrate atmospheric chemistry models.

Atmospheric-methane source and sink sensitivity analysis using Gaussian process emulation

We present a method to efficiently approximate the response of atmospheric-methane mole fraction and δ13C–CH4 to changes in uncertain emission and loss parameters in a three-dimensional global chemical transport model. Our approach, based on Gaussian process emulation, allows relationships between inputs and outputs in the model to be efficiently explored. The presented emulator successfully reproduces the chemical transport model output with a root-mean-square error of 1.0 ppb and 0.05 ‰ for hemispheric-methane mole fraction and δ13C–CH4, respectively, for 28 uncertain model inputs. The method is shown to outperform multiple linear regression because it captures non-linear relationships between inputs and outputs as well as the interaction between model input parameters. The emulator was used to determine how sensitive methane mole fraction and δ13C–CH4 are to the major source and sink components of the atmospheric budget given current estimates of their uncertainty. We find that our current knowledge of the methane budget, as inferred through hemispheric mole fraction observations, is limited primarily by uncertainty in the global mean hydroxyl radical concentration and freshwater emissions. Our work quantitatively determines the added value of measurements of δ13C–CH4, which are sensitive to some uncertain parameters to which mole fraction observations on their own are not. However, we demonstrate the critical importance of constraining isotopic initial conditions and isotopic source signatures, small uncertainties in which strongly influence long-term δ13C–CH4 trends because of the long timescales over which transient perturbations propagate through the atmosphere. Our results also demonstrate that the magnitude and trend of methane mole fraction and δ13C–CH4 can be strongly influenced by the combined uncertainty in more minor components of the atmospheric budget, which are often fixed and assumed to be well-known in inverse-modelling studies (e.g. emissions from termites, hydrates, and oceans). Overall, our work provides an overview of the sensitivity of atmospheric observations to budget uncertainties and outlines a method which could be employed to account for these uncertainties in future inverse-modelling systems.

Atmospheric methane source and sink sensitivity analysis using Gaussian process emulation

We present a method to efficiently approximate the response of atmospheric methane mole fraction and δ13C-CH4 to changes in uncertain emission and loss parameters in a three-dimensional global chemical transport model. Our approach, based on Gaussian process emulation, allows relationships between inputs and outputs in the model to be efficiently explored. The presented emulator successfully reproduces the chemical transport model output with a root-mean-square error of 1.2 ppb and 0.06 ‰ for hemispheric methane mole fraction and δ13C-CH4, respectively, for 28 uncertain model inputs. The method is shown to outperform multiple linear regression, because it captures non-linear relationships between inputs and outputs, as well as the interaction between model input parameters. The emulator was used to determine how sensitive methane mole fraction and δ13C-CH4 are to the major source and sink components of the atmospheric budget, given current estimates of their uncertainty. We find that our current knowledge of the methane budget, as inferred through hemispheric mole fraction observations, is limited primarily by uncertainty in the global mean hydroxyl radical concentration and emissions from fresh water. Our work quantitatively determines the added value of measurements of δ13C-CH4, which are sensitive to some uncertain parameters that mole fraction observations on their own are not. However, we demonstrate the critical importance of constraining isotopic initial conditions and isotopic source signatures, small uncertainties in which strongly influence long-term δ13C-CH4 trends, because of the long timescales over which transient perturbations propagate through the atmosphere. Our results also demonstrate that the magnitude and trend of methane mole fraction and δ13C-CH4 can be strongly influenced by the combined uncertainty of more minor components of the atmospheric budget, which are often fixed and assumed to be well-known in inverse modelling studies (e.g. emissions from termites, hydrates, and oceans). Overall, our work provides an overview of the sensitivity of atmospheric observations to budget uncertainties and outlines a method which could be employed to account for these uncertainties in future inverse modelling systems.