5 Spatial model
5.2 Methodology
5.2.1 Variables and data sources
Data on the annual average O3 concentrations from the 979 training sites were used as the
dependent variable in model-building. Some studies use a transformation (usually logarithm) of concentrations in an attempt to better approximate linearity of the relationship, to achieve a normal distribution of the residuals, and also to avoid negative predictions. Untransformed long term mean O3 concentrations were used here, as there was no violation of the criterion for normality of the
distribution. An additional advantage of using untransformed concentration data, as opposed to transformed concentrations, is that it is easier to interpret the relationship between concentration and the predictor variables because the relationship is directly additive, rather than proportional, as with logarithmic transformations. This is supported by theoretical arguments about how O3
production occurs.
The environmental variables offered as predictors in the LUR were described in Section 3.2.3. These data consist of land cover, road length by type, meteorology, altitude, topographical exposure (topex) and distance to sea, measured for different window zones around each of the monitoring sites. Table 5.1 includes a description of each predictor, window sizes, and the required direction of effect in the final regression model. As already mentioned meteorological conditions often exert the major impact on concentrations. Relevant meteorological factors include solar radiation, temperature, precipitation and wind speed, precipitation (Section 3.2.3.5). Temperature and solar radiation are important factors in determining rates of photochemical reaction, which increases as solar radiation intensifies and as air temperature rises. Precipitation plays a role in wet dispersion and also acts a proxy of cloud cover, both of which reduce O3 concentrations. Wind speed, on the
other hand, is more complex. While it has often been shown to have a negative association with O3
concentrations, largely perhaps because it encourages dispersion and dilution, in some cases its effect is positive. This mainly appears to be because it brings in O3 from other, enrich areas or by
promoting turbulence and the vertical mixing of O3 from higher layers in the atmosphere.
Topography is also an important determinant, both in its own right and through associations with meteorology and human activities. Higher concentrations of O3 tend to occur at higher altitudes
(Coyle et al., 2002), largely because of the more intense solar radiation, but also because mountainous areas tend to contain few emission sources of NO or other scavengers. Topex, or topographic exposure, defines the degree of openness of the terrain, in terms of the relative relief. It can be expected to have a positive association with O3 concentrations both because it implies
and because these areas are less likely to contain major emission sources. Areas of negative topex (i.e. depressions or valleys), on the other hand, will be more shaded and thus have reduced solar radiation, and potentially act as transport corridors and the focus for settlement, thereby increasing emissions of NO and other scavengers. Distance to sea is a proxy both for the likelihood of influx of O3 from the ocean (i.e. open sea), where O3 formation is enhanced (Caballero et al., 2007), and for
the influence of maritime climatic conditions.
Traffic is the major source of O3 precursors, especially NOx. It could be represented in a range of
variables used in LUR such as traffic intensity or counts on the roads. Due to the difficulty in getting these data for the whole of Europe, and following the example of many LUR studies, road length by type was used as a proxy for traffic volume. Several studies have shown that road length is a good substitute for traffic, giving similar results in LUR models (Vienneau et al., 2010, Henderson et al., 2007, Madsen et al., 2007).
Finally, land cover data are used to account for other sources of O3 precursors, especially from
anthropogenic and biogenic sources. For example, high density and low density residential lands are proxies for emission from domestic activities (e.g. heating). Industrial, commercial, construction, and port areas, were combined together in one variable, termed industrial/commercial land to provide a proxy for industrial emissions. Forest and agricultural lands can also produce O3 precursors,
especially VOC. However, their effects may be more complex. In the growing season, for example, some trees and agricultural crops may also act to diminish O3 concentrations by encouraging
deposition by absorbing O3 through the stomata of vegetation (Coyle et. al., 2002).
Some LUR studies for other pollutants also used population as a predictor variable (Skene et al., 2010). If the LUR model is to be used in an epidemiological study with an ecological design, or in a health impact assessment for the whole population, the use of population as a predictor can cause difficulties, for it results in a degree of duplication. It was thus decided a priori that population would not be used in this LUR model for O3. In any event, good land cover data will usually provide
an adequate and higher resolution measure of population distribution.
All of the above mentioned data sets are available across the whole study area. The regression equation thus developed from the LUR model can therefore be used to predict O3 concentrations at
unmonitored locations (i.e. all 100 metre grid cells) across Western Europe for mapping purposes. A summary of the available data is provided in Table 5.1.