Site-Specific Weather Forecasts for Site-specific Urban Air Quality Prediction

State-of-the-art mesoscale numerical weather prediction (NWP) models presently operate at scales, which only resolve the crudest effects of the largest of urban areas. ‘Added value’ forecaster intervention is required to more accurately estimate the urban impact of local and downwind weather conditions. At the Second Urban Environment Symposium in Albuquerque, Best (1998) and Clark (1998) presented details of a site-specific forecast model (SSFM) developed at the UK Meteorological Office (UKMO). This model attempts to simulate boundary layer structure more accurately by taking account of upwind roughness characteristics. The SSFM is coupled to the UKMO Mesoscale NWP model to provide forecast input boundary conditions. The initial applications of this work were more accurate forecasts of road surface temperature and airfield weather conditions.

Also at the Albuquerque meeting, Dorling and Gardner (1998) presented research which utilised neural network based models to retrospectively simulate urban air quality variability. This work highlighted the significant non-linear effects of local weather conditions on air quality and stressed the significance of maximising the accuracy of weather forecasts if this approach were to be adopted in forecast mode. Funding from the Leverhulme Trust is allowing this work to be extended to the operational air quality forecasting phase. This research is now ongoing in collaboration with UKMO, coupling the SSFM and neural network methods.

Initial results will be presented comparing the relative accuracies of the SSFM, standard Model Output Statistics (MOS) procedures, and Neural Network (non-linear regression) methods in reproducing urban weather conditions at weather station locations in a range of UK urban settings. The relative performances of the different approaches are likely, themselves, to be weather dependant and partly a function of the geographical position of the observing station relative to the urban area itself and the prevailing airflow. Therefore, in terms of the adoption of an approach which most accurately forecasts urban weather conditions on which neural network based air quality forecasts could be based, a hybrid approach is anticipated. A generic system will be aimed for as far as is possible in order that the approach can be ported to other potential study areas. This work links heavily to an ongoing European Union funded project (APPETISE) in which a model-intercomparison is being undertaken of a variety of statistically based air quality modelling approaches.