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

摘要

Sensitivity analysis of atmospheric models is necessary to identify theprocesses that lead to uncertainty in model predictions, to help understandmodel diversity through comparison of driving processes, and to prioritise research. Assessing the effect ofparameter uncertainty in complex models is challenging and often limited byCPU constraints. Here we present a cost-effective application ofvariance-based sensitivity analysis to quantify the sensitivity of a 3-Dglobal aerosol model to uncertain parameters. A Gaussian process emulator isused to estimate the model output across multi-dimensional parameter space,using information from a small number of model runs at points chosen using aLatin hypercube space-filling design. Gaussian process emulation is aBayesian approach that uses information from the model runs along with someprior assumptions about the model behaviour to predict model outputeverywhere in the uncertainty space. We use the Gaussian process emulator tocalculate the percentage of expected output variance explained by uncertaintyin global aerosol model parameters and their interactions. To demonstrate thetechnique, we show examples of cloud condensation nuclei (CCN) sensitivity to8 model parameters in polluted and remote marine environments as a functionof altitude. In the polluted environment 95 % of the variance of CCNconcentration is described by uncertainty in the 8 parameters (excludingtheir interaction effects) and is dominated by the uncertainty in the sulphuremissions, which explains 80 % of the variance. However, in the remote regionparameter interaction effects become important, accounting for up to 40 % ofthe total variance. Some parameters are shown to have a negligible individualeffect but a substantial interaction effect. Such sensitivities would not bedetected in the commonly used single parameter perturbation experiments,which would therefore underpredict total uncertainty. Gaussian processemulation is shown to be an efficient and useful technique for quantifyingparameter sensitivity in complex global atmospheric models.