Reducing uncertainty in emission estimates using perturbed emissions ensembles and novel observations: A focus on Beijing

The time lag between the occurrence of emissions and the compilation of an inventory is inevitable. When an emissions inventory is used to simulate air quality, uncertainties in the emissions are propagated into uncertainties in the modelled pollutant concentrations. Such uncertainties can be particularly high in regions undergoing rapid emission changes. Beijing, for instance, has implemented a series of pollution control measures over the past several years and various studies have confirmed significant decreases in the emissions of pollutants such as CO and NO_X. Hence, it is crucial to quantify and constrain the uncertainties in existing emission estimates for this region.

We sample the uncertainties in an emissions inventory for Beijing using a high-resolution advanced Gaussian dispersion model with perturbed emissions ensembles (PEEs), and constrain these uncertainties using a comprehensive set of in situ observations, including vertically resolved measurements made from a tower in central Beijing using low-cost sensors. We first construct a PEE by varying key emission parameters including source sectors, vertical and diurnal profiles within their uncertainty ranges estimated through expert elicitation. By removing the baseline contribution to the concentrations, we are able to evaluate the performance of the PEE in simulating the local signal. Based on knowledge gained from the initial PEE, we design a second PEE with optimised uncertainty ranges with which we constrain the uncertainties in the base emission estimates.

Our study shows the applicability of perturbed emissions ensembles and high-resolution, three-dimensional observations in systematically sampling and constraining emission uncertainties. This method has wide implications for air quality modelling, particularly in regions with rapid emission changes or for studies in which emissions inventories are out-dated.