Search

Air pollution is the primary environmental cause of premature death in the EU (European Commission, 2013) and the most problematic pollutants across Europe have consistently been oxides of nitrogen (e.g. nitrogen dioxide (NO2)), particulate matter (e.g. PM10, PM2.5) and ozone (O3). While measurements form an important aspect of air quality assessment, on their own they are unlikely to be sufficient to provide an accurate spatial and temporal description of the pollutant concentrations for exposure assessment and moreover they cannot provide information regarding future air quality. Annex XVI of 2008/50/EC requires member states to “ensure that up to date information on ambient concentrations of the pollutants covered” by the Directive are “made available to the public”. This information must include actual or predicted exceedances of alert and information thresholds and a forecast for the following day of which a model is an integral part. As a result, air quality models are increasingly required for public information, air quality management and research purposes. The primary objectives of this research fellowship were to develop a calibrated air quality forecast model for Ireland capable of predicting the Air Quality Index for Health (AQIH) in each of the air quality zones in Ireland and to model the spatial variation in concentrations on a national scale. This research project has produced three different models for NO2, PM10, PM2.5, O3 and SO2, all of which are available for further use. These are: A hybrid point wise 48 hour forecast model; Spatial model (WS-LUR) to produce annual mean maps of air pollution on a national scale; Temporal WS-LUR model. A comprehensive review of modelling systems carried out at the outset of this research fellowship, together with consideration for key EPA objectives, informed the direction of model development. This review is available as a separate EPA report. A priority within the EPA was to produce air quality forecasts based on the AQIH. The AQIH is based on point wise measurements and in order to extrapolate these measurements to the future, statistical modelling was deemed the most suitable. The advantages of this approach were that it could be developed from first principles specific to the area of interest and completely (avoiding any reliance on a third party to supply the model or apply licensing restrictions) and the associated speed of forecast computation. Forecasts are only useful if they can be computed and made available to the public relatively quickly. The accuracy of such methods also tends to be high and of low bias as they are developed site-specifically unlike large scale deterministic models that are often developed and tested in vastly differing domains. In particular, this method was capable of producing accurate point wise forecasts without the need for a detailed emissions inventory. At the project outset, the emissions inventory was not of sufficient spatial resolution to make realistic point wise forecasts in all air quality zones by deterministic means and it would have been an inefficient use of resources to base the development of forecasts on what was currently available. Initial model development proceeded using time series analysis in conjunction with non-parametric kernel regression, with local meteorological parameters as predictor variables. A model validation study found that this technique produced accurate forecast of ozone and SO2 but had a tendency to under predict peak NO2 and PM10/2.5 concentrations. An analysis of air mass history using the HYSPLIT model was carried out which revealed certain air masses (primarily easterly and re-circulated air) were responsible for most incidence of elevated concentrations. The results of this study were used to develop a HYSPLIT add-on for the forecast model which operates by forecasting air mass history in real time and invoking a different forecasting methodology depending on the region of origin of the air. The ability of the hybrid point wise model to predict daily maximum hourly NO2, SO2, 8 hourly ozone and daily average PM10 and PM2.5 was demonstrated by comparing a full year of modelled data with measured data at each of the AQIH sites. Index of agreement values ranged from a low of 0.80 for SO2 to 0.88 for NO2 and ozone, while correlation coefficients ranged from a low of 0.69 for SO2 to 0.82 for NO2. Full results of this validation study are contained in a separate report. In order to provide detail on the spatial variation of concentration levels across the country, land use regression (LUR) was recommended in the model review as the most suitable technique. This technique uses surrogate spatial indicators to explain the variation in concentration levels between monitoring points. Land cover data (CORINE), DTM output, road density information and population data are all factors that influence concentration levels and data that were broadly available. In contrast to most LUR studies, circular buffers were not used in the determination of spatial predictor variables. Rather, a novel sector based approach (WS-LUR) was adopted whereby variables were calculated within 8 pre-defined sectors representing the major wind directions around each monitoring site. This approach had a dual purpose. Firstly, it accounts for the direction of influence of emission sources on air quality in a given location. Traditional LUR assumes equal influence of emissions in the area surrounding a monitoring site regardless of wind direction. This approximation may be reasonable in highly urbanised areas where emissions sources are relatively uniform in the surrounding region. However, in this study the regression was applied on a national scale and prevailing winds coupled with clear directional influenced at air quality monitoring sites meant that WS-LUR is a superior option. The second advantage of this methodology is that it increases the effective number of data points available for the regression analysis, resulting in a more robust final equation. In conjunction with research project (2013-EH-FS-7), a set of annual mean maps within a geographic information system (GIS) environment were created and validated for each of NO2, PM10, PM2.5, O3 and SO2. These provide a highly relevant source of information regarding spatial variation in concentration levels on a national scale which can be used not only for exposure studies and general air quality assessment, but also as a tool to correlate emission sources and surrogates with air quality. A temporal WS-LUR model was developed for NO2, Ozone and PM10 by including hourly meteorological data in conjunction with pre-specified spatial data as predictor variables. This model has the potential to provide fast, efficient national air quality forecast maps for Ireland with minimal computational requirements. This project has achieved key EPA objectives and has produced a fully automated and operational air quality model which produced twice-daily forecasts of the AQIH in each air quality zone in Ireland. The stepwise approach chosen for model development allowed deliverables prior to completion of the project while minimising associated risks. The models developed as part of this fellowship form solid building blocks on which future air quality modelling studies in Ireland can be based.