Urban emission inventory optimisation using sensor data, an urban air quality model and inversion techniques

Abstract

An optimisation scheme has been developed that applies a Bayesian inversion technique to a high resolution (street-level) atmospheric dispersion model to modify pollution emission rates based on sensor data. The scheme minimises a cost function using a non-negative least squares solver. For the required covariance matrices, assumptions are made regarding the magnitude of the uncertainties in source emissions and measurements and the correlation in uncertainties between different source emissions and different measurement sites. The scheme has been tested in an initial case study in Cambridge using monitored data from four reference monitors and 20 AQMesh sensor pods for the period 30 June 2016 to 30 September 2016. Hourly NOx concentrations from road sources modelled using ADMS-Urban and observed concentrations were processed using the optimisation scheme and the adjusted emissions were re-modelled. The optimisation scheme reduced average road emissions on average by 6.5% compared to the original estimates, changed the diurnal profile of emissions and improved model accuracy at four reference sites.