Bad NO2

How can the NO2 measurements go badly wrong?!

Here is a simple example during an episode. Remember that the NO2 air quality standard is based on the number of hourly mean exceedences of 200 ug m-3 or nearly 105 ppb. You may think that low concentrations are the hardest to measure and these are not important in urban areas. However, episodes are the crucial periods and this is when the NO2 data processing errors are the greatest!

Imagine the real ambient hourly mean concentrations are

NOx = 1,000 ppb, NO = 900 ppb and NO2 = 100 ppb which is not an exceedence.

Make a small +1% error on the NOx channel calibration and the measurements now become

NOx = 1,010 ppb, NO = 900 ppb and NO2 = 110 ppb which is an exceedence.

The tiny +1% error on the NOx can easily become a +10% error on the NO2. Similarly a -1% error on the NO will produce the same skewed result. This tips the NO2 measurement into an exceedence.

Here is a real example with the hourly mean NO2 peaking at 130 ppb and NO 450 ppb. There were 5 exceedences of the NO2 104.6 ppb hourly limit value.

The effect on the NO2 concentrations of a small +1% error on the NOx and -1% NO sensitivities may appear small. However, the number of NO2 exceedences has now doubled from 5 to 11. Note that the effect increases with NO2 concentrations and is worst during an episode when correct monitoring is most important. Exceedence counts are very sensitive to small changes near to the limit threshold.

Where does the 1% error come from?

Most organisations scale the data with constant sensitivity factors until the next calibration. These steps will not follow a drifting instrument very well. The best method is a ramp between the calibration points.

This plot shows the drifting NOx sensitivity over two months. The real drift is shown in blue. The site operator is calibrating the instrument every fortnight and applies a constant sensitivity (green line) until the next calibration. Clearly there is an error shown in red between the site operator's estimate and the real sensitivity. Note this method produces significant errors even during periods of modest drift. The worked example above shows that tiny 1% errors can be disastrous during an episode.

Applying a constant sensitivity is common practice. Actually plotting the sensitivity drift is a novelty for many site operators. Normally this makes little difference but during episodes, the most important period, the reported NO2 concentrations can be wildly wrong. The best way to scale NOx measurements is by a series of ramps between the calibration points that closely follow the natural instrument drift. This technique is shown in the plot below. Errors are now less than 0.5% even during periods of rapid drift. This operation is difficult to perform in a spreadsheet but is very routine in specially designed software. This is the method I pioneered in the AURN and have always used since 1988.

Here are some common examples of bad practice

• Resetting the instrument to the calibration at every visit. The data processing is very easy because there isn't any! The instrument is wrongly assumed to read correctly. This practice is effectively the same as applying a constant sensitivity as shown above. However, the errors can be far worse. What happens if the calibration system was leaking, the span was performed too quickly or there was no excess flow? Correcting for these common problems is impossible since the instrument was adjusted to an unknown calibration.
• NO calibration cylinder concentration certificates. Most concentration certificates say something like NO 400 ppb with NO2 < 5 ppb. This is wrongly assumed to mean NO 400 ppb, NOx 400 ppb and NO2 0 ppb. The cylinder may really be NO 400 ppb, NOx 404 ppb and NO2 4 ppb. This is a 1% error on the NOx concentration. Not much can be done while the cylinder suppliers provide these uncertain certificates.
• Only collecting the NO2 readings from the instrument. This is wrong since the NO2 instrument channel is not directly proportional to the ambient NO2 concentrations. This is shown in an example NO2 calculation spreadsheet. The NO2 channel cannot be calibrated with a NO cylinder so this channel should never be used.
• Only processing the NO2 concentrations. Very often only the NO2 is examined without any regard to NO or NOx concentrations. The site operator does not know that the NOx instrument only measures NO and does not actually measure NO2. The NO and NOx concentrations must be processed first to calculate NO2. Neither the instrument or the software will do this without some manual expertise.
• Believing every calibration. Mistakes are easy to make and sometimes the instrument has faults. Calibrations should be plotted and bad values ignored.
• Believing the zero calibrations. Spent zero scrubbers allow through ambient NO2 but easily remove all ambient NO. The NOx calibration zero can becomes a small 4 ppb while the NO remains 0 ppb. This then leads to a 1% error on the NOx if the span is around 400 ppb. A calibration plot will show the NOx zero wandering while the NO zero remains stable.
• NO calibration cylinders can become progressively poisoned by oxygen. The NO concentration can often halve while the NOx remains relatively unchanged. This is obvious if the calibrations are plotted but can produce enormous errors if ignored.
• Bespoke spreadsheets that do not work. These include spreadsheets where NOx does not always equal NO + NO2. Concentrations reported to 16 decimal places. Negatives simply replaced by 0 ppb. NO2 concentrations when the NO is missing. Missing measurements stored as 0 ppb. NO and NOx ppb not shown. No way to plot the calibration drift. No way to plot concentrations for more than one month.
• The Rogues Gallery shows some real examples of obviously bad data processing.

A quick look at the NO concentration plot often shows if the NO2 data processing has been performed correctly. However, NO is rarely reported or even calculated! Occasionally some NO2 peaks and dips look odd but without the sensitivity plots you just have to believe the concentrations.

Most reported NO2 data are just the annual mean and exceedence statistics. No one knows if these important NO2 statistics are based on simple data processing using sound methodologies or are just a waste of time and resources.

Remember a tiny 1% error can easily overestimate the NO2 concentrations by 10% !!

How many AQMAs have been declared or not declared based on suspect NO2 data?

Don't rely on your instrument or your software to produce reliable concentrations without some manual expertise. NOx instruments drift, that is a fact of life, and you must track this drift closely or suffer anomalous peaks and dips.

You need an expert to check, select, smooth and apply the best calibration scalings to your measurements.

Contact me Geoff.Broughton@aqdm.co.uk