This section describes the first stage of applying EHIPS to data on the city of Newcastle, New South Wales, Australia. The data include several industrial emission sources and measured meteorology and pollutant concentrations at three sites: Newcastle proper, Wallsend (westward) and Beresfield (north-westward). The EHIPS team's thanks for access to these data are due to the following persons:
Prof. Howard Bridgman, Newcastle University
Olivier Rey-Lescure, Department of Geography, Newcastle University
Dr Rhonda Boyle, EPA, State of Victoria.
We plan the following sequence of activities for applying EHIPS to Newcastle data.
Stage 1: technical – plugging in and checking the databases, making the map/GIS base, checking for units and consistencies, obtaining first data overviews.
Stage 2: concentrations – integrating the model calculations of dispersion from sources and the actual measurements. Localizing/customizing the model choice and model parameters for use at subsequent stages. Establishing basic concentration patterns across time/space/pollutant.
Stage 3: risks – establishing a sensible local model of exposure groups and exposure scenarios. Identifying basic risk patterns, especially hotspots across time/space/exposure group.
Stage 4: health – calculating expected patterns of morbidity/mortality, validating them by actual health data, searching for statistically significant relationships between health and risk patterns, identifying mutually supporting environment and health hotspots.
Each stage is a necessary prerequisite for the next one.Presently, we are finishing Stage 1. As a result of this stage, we have produced various data overviews, some of which are shown below. Model calculations have been performed using the simplest possible configuration of dispersion modeling:
Both the US EPA model ISC3ST and the Russian model OND-86, with extensions made specially for EHIPS, were used;
Terrain or buildings were not taken into account;
Special phenomena, such as dry or wet deposition, mixing height variation etc. have not been taken into account;
Atmospheric stability conditions were taken into account through the regional coefficient, as usual in OND, and through the wind speed variations sigma-theta in ISC;
Low-level and area sources were temporarily dropped out because they shouldn't influence much the far-off receptors where actual concentrations are measured.
Therefore, these calculations are good only for data overview and error screening, as it should be at this stage. Additionally, it was important to assess the discrepancy between the model and measured concentrations.
Our plans for the Stage 2 (it is too early to speak about later stages) include:
Elaborating a varying-detail system of spatial cells for dispersion modeling. It should be adapted to map configuration, to source/station position, and to population reporting unit system (POs). All further calculations will be done on this system of cells.
Taking into account all phenomena dropped before (stability parameter is discussed separately below).
Comparing ISC and OND models and making a decision on using one of them or both concurrently, within some information sharing scheme.
Choosing one of several existing methods for calculating the atmospheric stability parameter. The method chosen should be applicable both to ISC and OND, unless it is decided to abandon one of these models outright. The major criterion of method choice (as well as of model choice above) will be the consistency between modeled and measured data (values or patterns). The model fitting capabilities of EHIPS will be used.
Identifying concentration patterns characteristic for industrial emissions and trying to recognize them in measurement data. This could be used to assess the proportion of unaccounted-for emissions (mobile etc.) and to improve the quantitative agreement between modeled and measured concentrations.
Finally, the modeling results and measurements should be merged into a single concentration field over the cell system mentioned above. These concentrations will be used throughout the subsequent stages.
What follows is the data overview. We have divided it into the following sections:
Figure 1 This is the geographical substrate of data overview. Thin violet lines delimit PO's and thick lines, districts. Small squares with green icons are sources (stacks only) – approx. 100 of them. Small squares with blue icons are stations: top left – Beresfield, bottom right – Newcastle, Wallsend is hidden behind one of the cells. Big squares with crosses are cells into which we have split the territory for dispersion modeling. For technical reasons some of them were not shown.
Figure 2 This is how a typical dispersion modeling result looks when shown as a map. Color codes are for calculated concentrations of a substance. Their numerical values are shown at the top. White normally means no data, but high values may be also shown as white if above the fixed high limit (that is, if no automatic normalization is used for colors). Note that calculations were done both for cell positions and for station positions, so both are colored. PO territories are filled by color corresponding to the average value of cells overlaying them.
This sample map was obtained for a single time step – an hour.
Figure 3 The same type of result with cells removed so that POs are visible better. This was obtained by averaging the hourly CO concentrations calculated for a year's length. All values for color codes shown at top are in milligrams per cubic meter – here and henceforth.
Figure 4 The same for NOx. While the general picture is similar to Figure 3, the values are lower than for CO by approximately a factor of 5. Also, note that the spatial pattern of pollution is somewhat less extended.
Now, several maps characterizing the emission source parameters.
The single predominant source of emission for both CO and NOx is a group of stacks belonging to the enterprise encoded as N4. The situation is different for SO2 as seen in the graphs below.
The graphing module of EHIPS shows data as time series or rankings or histograms. Rankings are useful for selecting the most important sources.
Figure 8 Ranking of stacks by emission of NOx. Two stacks are prevalent, from three to fifteen more are more or less important, and the impact of remainder could be analyzed for possibility of screening them off in order to accelerate calculations.
We give this picture also as an example of the graphing module screen interface.
Figure 9 Ranking of stacks by emission of SO2. Again, several stacks are dominant, but they are different.
Figure 10 Ranking of stacks by temperature. This is an important indicator of the possible range of a stack's impact (due to buoyancy).
Figure 11 Ranking of stacks by height. This is an indicator of the perceived danger of emissions. Note a much more equal distribution than for T.
Measured values
This is the overview of a year's worth of measurements made at 3 stations (positions shown below).
Figure 12 Yearly averages of ozone measurements at three stations. Note that Newcastle values are as low as Beresfield values, although Newcastle is seemingly in a more dangerous position w.r.t major emission sources. This is true for most other pollutants, as well. So it's, probably, a result of a specific combination of meteorology and position. This can be checked by modeling.
Figure 13 A graph of yearly averages for ozone at three stations. Newcastle is again as low as Beresfield.
In the following figure, the concentrations at each station are unfolded by month. (Note that the vertical scale for Wallsend is twice as high as for Newcastle and three times as high as for Beresfield. However, this is due to a single 'high' month, and yearly averages are much closer. Note a much more pronounced seasonal pattern for Newcastle (while, again, keeping a close tie on the average).
Figure 14 Beresfield: Monthly concentrations of NOx
Figure 16 Newcastle: Monthly concentrations of NOx
Now, let us compare the patterns of yearly average concentrations as unfolded across pollutants. (Note that not all pollutants are measured at all stations. Note also the difference in vertical scale.) The pollutants go from left to right as follows: CO, NO, NO2, NOx, O3, PM10, SO2.
Figure 17 Beresfield: Yearly concentrations by pollutant
Figure 18 Wallsend: Yearly concentrations by pollutant
Figure 19 Newcastle: Yearly concentrations by pollutant
At the surface level, patterns seem the same. So, we conclude that, in order to identify distinguishable patterns (see the Description section), pollutant and territory dimensions alone may be insufficient, and the time dimension should be taken into the bargain.
A convenient method for visualizing the values fluctuating in time is their histogram. Below, we show several full-year histogram of hourly measurements.
Note a considerable difference in histograms for two related substances measured at a single station: NO has a much longer tail than NO2. As meteorology is the same, this is probably due to a difference in time variation of emissions.
When comparing histograms at different stations, it is important to recognize the influence of meteorological variations. These can be represented by histograms, as well. Below, we give an example for wind speed (again, hourly values for a year). The horizontal scale was kept comparable for all three (extending approximately to 10 m/sec).
Note that the high-speed tails are approximately the same, but there is a considerable difference in the low-speed part. It is important to recognize it, in order to put the right meteorology into modeling, especially because the low-speed limit is difficult for modeling and may produce unexpectedly high concentrations from otherwise less important stacks.
For time unfolding, even more important than the wind speed, is the wind direction. Its seasonal changes may determine the basic pattern of seasonal concentration change at a given station. Below, we give an example for Beresfield. First, we show the wind direction histograms for each second month (taken at Newcastle, but the seasonal pattern is the same), and then, the monthly histograms of measured hourly concentrations of NO at Beresfield. It is, in our opinion, quite evident visually that the high-value tails of concentrations are correlated to wind direction histograms biased toward right. Of course, it is sufficient to look at the map to see why it should be this way. However, histogram relationships (generally, pattern relationships) can be hoped to work also in absence of an evident cause-effect mechanism like that.
Figure 25 Beresfield: wind speed histograms by month. Note the difference in predominant winds between winter and summer. The plume extends toward Beresfield when the wind direction is close to 300 (see Figure 2).
Figure 26 Beresfield: histograms of hourly measured NO concentrations by month. The colors were made the same as in Figure 25. Note the obvious correspondence between the wind direction and the high-value tail of concentration histogram.
Model calculations using OND
First, let us compare the histograms of model concentrations calculated for positions corresponding to three stations.
Figure 27 Beresfield: Histogram of hourly model values for CO
Figure 28 Wallsend: Histogram of hourly model values for CO
Figure 29 Newcasl: Histogram of hourly model values for CO
Model concentrations are by an order of magnitude higher at Beresfield than at Newcastle and again several times higher at Wallsend. Measured Newcastle values are somewhere between modeled Wallsend and Beresfield values; unfortunately, no direct comparison is possible since CO was not measured at Wallsend or Beresfield.
Note also very expressive histogram patterns – they carry potential information on interaction between wind parameter change and source-to-receptor geometry.
This effect is lost on averaging. See the following monthly averages of calculated NOx values for positions corresponding to three stations. The seasonal pattern is insensitive to station position.
Figure 30 Beresfield: Monthly averages of model values for NOx
Figure 31 Wallsend: Monthly averages of model values for NOx
Figure 32 Newcastle: Monthly averages of model values for NOx
The absolute values change between stations in the same way as above for CO.
Now, let us compare the calculation and measurement by plotting the ratio of the two taken for corresponding points in space and time. Let's begin with NOx. The following are the yearly histograms of the hourly values of this ratio.
Figure 33 Beresfield: Yearly histogram of model values divided by measured values for NOx
Figure 34 Wallsend: Yearly histogram of model values divided by measured values for NOx
Figure 35 Newcastle: Yearly histogram of model values divided by measured values for NOx
The forms of the histograms are similar, and this suggests that the mechanisms of meteorological parameter variations are more or less correctly taken into account by the model, so they are factored out for each station. (If they were not, there should be differences between histograms as in Figure 27 - Figure 29). However, the average values are much lower in the model than in the measurement, as shown by the following figure.
Model calculations using ISC
Figure 36 Patterns of calculated concentrations by pollutant for different receptor territories corresponding to stations (furthest down the picture) and POs (the remainder). While all patterns are qualitatively similar, there are marked quantitative differences between territories. The pollutants go along the abscissa in the following sequence from left to right: CO, NO, NO2, NOx, O3, PM10, SO2, benzene, toluene, formaldehyde, xylene, Cd, Ni (the last four are meteoparameters).
The following histograms should be compared row-to-row. For characteristic pollutants such as CO and NOx (and also SO2, PM10 etc., for that matter), the histograms are long-tailed, so that short-lived intensive hazards are a phenomenon to be considered. For metals (Cd and Ni taken as examples), the histograms are much closer to normal distributions, so that an average, e.g., yearly, value can be a valid first-order approximation. This ties perfectly to the well-known mechanism of dose averaging for heavy metals' impact on health, in contrast to importance of acute exposures for CO, SO2 etc.
In both cases, the longest tail is for PO2304, which is the territory where most stacks belong.
Figure 37 Histograms of calculated CO concentrations split by territories.
Figure 38 Histograms of calculated NOx concentrations split by territories.
Figure 39 Histograms of calculated Ni concentrations split by territories.
Figure 40 Histograms of calculated Cd concentrations split by territories.
These are the spatial patterns of each pollutant concentration divided by concentration of NOx over the same area (taken as a reference value). Colors are normalized so that the PO with the maximal value is red. CO2 pattern is unique and the most extended one (note, e.g.,. the coloring of the rightmost and top left areas). All patterns have one thing in common: the core of top-value areas. This is due to the fact that these areas have low calculated concentrations for NOx taken as a reference. This core is more pronounced for toluene and xylene. As a first approximation, all maps shown can be taken as representing the same pattern.
The same data, with Ni concentrations taken as a reference. In our opinion, patterns clearly split into several groups. One is CO, PM10 and NOx. Another is formaldehyde alone. The third one includes benzene, toluene and xylene. The fourth one is represented by Cd. Obviously, this grouping has some relation to the chemical composition of pollutant. That's because related chemical are emitted by the similar groups of stacks.
Note that the pattern grouping depends on the choice of the reference.
Model calculations to measured values
Figure 57 OND model values divided by measured values for NOx. Yearly average for each station. (Zero columns named 'Meteo' are accessories to calculation – don't pay attention to them.)
The discrepancy may be due partly to unaccounted-for emission sources (e.g., mobile), and partly to modeling simplifications described in the beginning. A major task for the next stage is to distinguish between these two factors.
An argument for the importance of source factor is in the following figures showing the model-to-measurement histograms for SO2 (no measurements were made at Newcastle).
Figure 58 Beresfield: Yearly histogram of OND model values divided by measured values for SO2
Figure 59 Wallsend: Yearly histogram of OND model values divided by measured values for SO2
The difference in form is illusory – due mainly to difference in horizontal scale. As for averages, they are much closer to the desired target – that is, unity – than for NOx (see Figure 60). Maybe, that's because SO2 is more industry-specific than NOx, so its sources are more or less completely reflected in the model.
Figure 60 OND model values divided by measured values for SO2. Yearly average for each station. (No measurements available for Newcastle.)
