A multi-year dataset of measurements of CO2 concentrations, eddy covariance fluxes, and meteorological parameters over the city centre of Florence (Italy) has been analysed to assess the role of anthropogenic emissions and meteorology in controlling urban CO2 concentrations. The latter exhibited a negative correlation with air temperature, wind speed, solar radiation, and sensible heat flux and a positive one with relative humidity and emissions. A linear and an artificial neural network (ANN) model have been developed and validated for short-term modelling of 3-h CO2 concentrations. The ANN model performed better, with mean bias of 0.58 ppm, root mean square error within 30 ppm, and r2=0.49. Data clustering through the self-organized maps allowed to disentangle the role played by emissions and meteorological parameters in influencing CO2 concentrations. Sensitivity analysis of CO2 concentrations revealed a primary role played by the meteorological parameters, particularly wind speed. These results highlighted that (i) emission reduction actions at local urban scale should be better tied to actual and expected meteorological conditions and (ii) those actions alone have limited effects (e.g. a 20% emission reduction would result in a 3% CO2 concentrations reduction). For all these reasons, large-scale policies would be needed.
Keywords: Artificial neural network; CO2 concentrations; Eddy covariance; Meteorological conditions; Self-organized maps; Urban CO2 fluxes.