«Sef van den Elshout October 2014 AQI & AQ Communication D2.3/v07 PartII Work Package 2, Task 2.2 Deliverable D2.3 Version number V07 Ver Date Author ...»
33.0 32.0 31.0 30.0 29.0 28.0 27.0 26.0 25.0 Figure 1: Presentation issues with daily PM values On the left: hourly PM concentrations versus 24-hour moving average. The latter is flat line while the former clearly shows peaks related to the morning rush-hour and the built-up towards the evening rush-hour. On the right hand side: new year’s eve fireworks: the hourly data shows a short sharp peak, the moving average shows a moderate increase, that stays elevated long after the hourly concentrations have returned to normal. The 24-hour moving average data often display wrong (wrongly timed, too high or too low concentrations).
The US use both 8-hour and 1-hour O3 AQI calculation grids for those areas where the shorter averaging time is more appropriate/more informative. Likewise corresponding hourly and daily calculation grids for PM were developed for the CAQI.
1.3.3. Spatial representativeness Most information on AQI-s deals with the calculation grid and when trying to find information on the spatial representativeness of the AQI it is usually hard to find. However it is a detail that merits careful thought. An AQI can be calculated for an (administrative) area such as a city or a region. In that case area averaging has to take place. If this is not done and AQI-s are reported for individual monitoring sites it is left to the receiver of the information to decide what is the appropriate monitoring station providing information that is relevant for him or her. If one is in an area with little or no local gradients the nearest station is the most representative.
However if local sources dominate the occurring concentrations it highly depends on whether or not one is exposed to these local sources. This is information the lay public might not have/might not understand. It is one of the reasons why the CITEAIR project characterises a city by two indices: one for people in/near busy roads and one in the city background. If someone lives in a city background situation but the nearest monitoring site is a traffic station, it is better to disregard that information and rather look at a background site even if it is further away. Doing no averaging – even if one seems to provide more information in that case – might not be the best (the most informative) approach.
There are several ways to handle area averaging:
The easiest is not to average and report each station. The disadvantages are as mentioned: it still leaves the public with a lot of information to digest and interpret, and it makes it hard to compare one city to another.
In the US they report one index figure per city, this is also common in many other countries. In this case one has to decide if the city average is used or the highest value.
If averages are used one has to decide if the pollutant concentrations are averaged before the AQI is calculated or if the AQI per station is calculated and subsequently the AQI-s are averaged. If the highest value is used - at least for the AQI-s where the highest iAQI determines the AQI - the highest iAQI at any given time determines the index. On the other hand, if a time series of AQI-s is presented the AQI can alternatively be based on different monitoring sites making the time series harder to interpret.
Averaging can be done per pollution/exposure type, e.g. averaging for traffic exposure, for urban background, for industrial zones, etc. This makes it easier to compare situations/cities as the result is less dependent on the local monitoring strategy.
Other approaches are special cases of the three mentioned above. They mainly involve some kind of weighting e.g. by population numbers. In my view this makes sense for year average AQI-s that are used for policy monitoring. See section 3.1. It was proposed by Žujić et al (2009) for real time monitoring as certain monitoring sites were in busier areas than others. I think that the averaging by type (previous bullet) is more useful. Even if population weighting is applied it still depends heavily on where the monitoring sites happen to be located and not on how the population is exposed and for which part of the population the number is relevant.
If spatial and temporal averaging takes place one has to be clear on the order in which this is done. In particular for the indices where the highest iAQI determines the AQI, and where the
highest AQI in an area determines the area AQI (e.g city AQI), the order in which things are done may impact the outcome and which pollutant determines the AQI. E.g. average every hour across all the monitoring sites for each pollutant and subsequently determine a daily average before determining iAQI-s or determine daily average iAQI-s at each site and subsequently determine the area AQI.
1.3.4. The choice of pollutants in the AQI The choice of pollutants to be included in an AQI varies considerably. Most provide calculation grids for ozone, nitrogen dioxide and PM10. Lately several AQI-s have added PM2.5. Other pollutants such as sulphur dioxide, carbon monoxide and benzene occur as well. To some extent the pollutant mix in the AQI depends on the air pollution history of the area. In Europe, the old monitoring networks dating back to the 1960ies and 70ies tended to have SO2 included, whilst the younger networks (dating from a time when industrial air pollution has largely been replaced by traffic related air pollution) typically included CO instead of SO2. Which pollutants to include should be based on the actual nature of the prevailing air pollution. As a general rule the index should include at least O3 and PM10 and/or PM2.5. Both are health relevant, occurring fairly universal, transported over large distances and wholly or partly formed in the atmosphere.
For local combustion sources including traffic, NO2 (or NOx, though I’m not aware of any AQI that uses it) is a suitable indicator and it appears in most if not all, AQI-s.
How to treat missing pollutants is an issue that is hardly ever treated explicitly. The fact that for most AQI calculations individual iAQI-s per pollutant are determined implicitly suggests that all pollutants are equal and carry a similar weight. It suggests that if one of the pollutants is missing one can still calculate the AQI, after all the highest available iAQI determines the AQI.
Technically this is correct but does this AQI based on only a few or even one pollutant have the same meaning?
The CAQI was specifically intended to make air quality comparable in different cities in different countries and for that purpose a minimum set of pollutants was specified to be able to calculate the CAQI. The developers felt that without a minimum set of common indicators comparability could not be assured (Elshout et al, 2013, 2008). The US AQI takes the view that every pollutant ‘has its own AQI’. Even if not all pollutants are available an AQI is presented. In one way this is a valid consideration: if one pollutant is high, one would want to alert the public and not withhold that warning simply for the administrative reason that the information is not complete. Not doing so would be a mistake.
On the other hand, if several pollutants are missing and the AQI is low providing that information one might mistakenly send a reassuring signal. In short, the selection of a minimum set of pollutants to be included in the index needs careful thought and analysis of available data. This also applies to the question of how to treat missing values of pollutants, and the need to adapt the messages accompanying the AQI, in this case.
For the health based AQI-s that take pollution interaction into account, a missing pollutant leads to AQI values that are too low5. This is obvious. For the AQI-s where the highest pollutant determines the AQI the same might happen. Practical results from the CAQI (see box 4) demonstrate this. Results from data analysis on Chinese data shows that in its current form both PM fractions would be needed. This is discussed more extensively in document part III.
Box 4: Sensitivity of an AQI to the pollutants included
The CAQI index requires that at least three pollutants are available to be able to calculate an AQI. The table presents the frequency of the distribution of each index class if the CAQI is calculated according to the existing calculation rules or if it had been based on one of the individual AQI-s. As can be seen this would have led to a completely different outcome.
For this sample of 31 European urban background monitoring stations at least the data for PM10 and O3 would have to be available (‘CAQI minimum set’ - blue column).
If one takes seasonal aspects into account the differences are even bigger. Leaving out PM for example leads to completely wrong wintertime AQI values. (Elshout et al, 2013).
Note that including pollutants that are strongly correlated could lead to AQI values that are too high in these additive health effects AQI-s. This depends on the way the Relative Risks were determined (single versus multi-pollutant models).
2. Recent developments – health based indices
2.1. Introduction Looking at the scientific literature there have been a number of publications on AQI-s. Most of them tend to stress the importance of a sound health base, e.g. Sicard et al. (2012), Chen et al.
(2013), Dimitriou et al. (2013) and Wong et al. (2013). The more sophisticated health indices
aim to address a few issues that exist with other AQI-s. E.g.:
a) If the AQI class is bad, or high, what does it mean? How bad is it, what can I expect?
b) If more than one pollutant is high, is there an extra risk?
c) If the AQI for PM10 is x and the AQI for NO2 is also x, do the two pose the same health risk?
Though many of these indices are published as they make interesting research material, few of them are actually used. In earlier sections we discussed that health based indices have several communication issues and the more sophisticated indices suffer from high complexity. Often index bands are explained in terms of relative risks or excess risks. Though these concepts are familiar to epidemiologists, they are very hard to understand for the lay-public. If the index is mainly used for exposure monitoring then these complex indices are the best solution.
Recently Hong Kong actually implemented a complex health based index taking pollutant interaction into account following Canada, where it was implemented as well. In these cases relative risk is used to determine index classes that can be communicated more easily to the public. To my knowledge Canada and Hong Kong are the only places where a sophisticated health based index is actually used.
In the Netherlands the existing air quality index is under review and is likely to be replaced by a simple health based index. In the Netherlands pollutant interaction won’t be taken into account but the above 3 questions are addressed in a simplified way.
In the next two sections these two recent developments are briefly presented. Some more info on the Canadian index is available in Annex A1.1.
2.2. The Hong Kong AQHI In Hong Kong the air quality and health index or AQHI is operational since 2014. It replaces an API (Air pollution index) that strongly resembles the US and Chinese AQI-s. The AQHI follows closely the Canadian index. Local epidemiological studies were conducted to study the Relative Risks associated to air pollution exposure. Hospital admissions were chosen as health endpoint.
SO2 was included in Hong Kong (not in Canada) to reflect local air pollution problems. The index now includes NO2, O3, PM10 and SO2.
The authors have also established relative risk factors for PM2.5 and this pollutant could have been included in the AQHI as well. However, preference was given to PM 10 because it is Page 17 of 43 AQ Communication D2.3/V7 part II monitored at more stations and it is more relevant during (coarse dust storms). Including both at the same time would exaggerate the risk because both species are highly correlated.
The total excess risk is calculated as:
Excess Risk = Σ((RRj-1)*Cj) The AQI is based on 3 hour moving averages. This is a good compromise between real-time information and getting stable results. Based on Hong Kong monitoring data the ER would range from 0 to 19.37%. This is then converted into 10 bands, each associated with a risk band, plus an additional band ‘10+’ covering even higher risks (concentrations), should they arise. The bands can be arbitrarily chosen (Canada) or partly linked to policy standards or WHO guidelines.
In Hong Kong they tied band 8 to the WHO short-term guidelines and arranged the rest accordingly. The highest band occurs rarely (2.7% of the time) which is excellent to avoid message fatigue. For more information see Wong et al (2013).
The AQHI addresses the three issues in 2.1 in the following way:
• Relative criteria such as ‘low’ … ‘high’ are tied to epidemiologically determined risks.
Though risk is a difficult concept for lay people, the approach provides a factual basis for the relative categories.
• Risk are used in an additive way. So if two pollutants are high the risk increases.
Whether it is fully correct to simply add the risk is somewhat doubtful. However, since the risks are only used to provide a science base for what is still a relative measure for the severity of the pollution, this approach is sufficiently correct.
• The AQHI doesn’t provide iAQI-s for individual pollutants so this question doesn’t arises with this AQI. At the fundamentals of the AQI calculation, the true relative risk per 10 µg/m3 for each pollutant is used. In that sense the AQI gives each pollutant the appropriate health weight.
The fact that there is no table with grids and that it is not easily possible to identify the pollutant that contributes most to the pollution risk at a given time is a (small) disadvantage from a communication and educational point of view. If this AQHI is meant to monitor public exposure over time it is a very good option.