Abstract/Summary

Air quality (AQ) forecasts are important for helping the public mitigate adverse health effects associated with episodes of high pollution. In order to create an accurate AQ forecast, it is important to correctly represent meteorological processes due to their strong influence on pollutant concentrations. The aim of this work is to determine the relationship between forecast errors in surface pollutant concentrations and meteorological forecast errors within the operational AQ model, Air Quality in the Unified Model (AQUM). This thesis explores three different approaches of evaluation to determine the impact of meteorological forecast errors in 10 m wind speed, 1.5 m temperature and precipitation on pollutant errors (O3, NO2, PM10 and PM2.5). These are point-based, neighbourhood and process-based methods. Point-based metrics evaluate forecasts against observations, paired in space and time. The evaluation reveals negative forecast bias in diurnal cycles of summertime NO2 and positive bias in O3, with a 2-hour lag in timing of the forecast increase of morning concentrations. It is shown that night-time 10 m wind speed over-estimation coincides with the largest O3 and NO2 biases. Point-based evaluation identifies a negative bias in PM concentrations which decreases by 10% to 25% after under-estimating precipitation. Neighbourhood evaluation relaxes the spatial constraint for forecast - observation pairs. It is a novel mechanism of attributing forecast errors in AQ to meteorology. Strongest positive correlations between night-time O3 and wind speed forecast errors are shown to occur at a neighbourhood of 100 km2 . Forecast error anti-correlations between NO2 and wind speed reach a maximum at a smaller neighbourhood. Finally, process-based evaluation is used to test whether statistical relationships between O3, NO2 and 10 m wind speed forecast errors are caused by the physical process of entrainment. To quantify the influence of entrainment in the model on forecast total oxidant (O3 + NO2 = Ox) concentrations, an experiment using the off-line NAME model simulates tracer dispersion under different meteorology configurations. The experiment confirms that lag in the forecast morning Ox increase is due to delayed boundary layer development. In light of the results, it is recommended that model developers implement a land-surface parametrisation with better urban heat storage to improve modelled surface heat fluxes, nocturnal boundary layer stability and nocturnal winds. Improving these may result in more accurate surface AQ forecasts.