Correction Algorithms

The accuracy of an air quality monitor is impacted by a variety of internal and external factors, for example:

• Enclosure: The ventilation and internal placement of sensor modules can greatly impact their accuracy.
• Relative Humidity: The relative humidity can increase the size of particulate matter.

Therefore, air quality device manufacturers compare the performance of their monitors with reference instruments and then apply correction formulas that could be as simple as a temperature offset but also involve complex multivariable models.

In most cases, manufacturers treat these algorithms as trade secrets and do not publish them. We at AirGradient believe in open hardware and open science and, therefore, openly publish them. This allows not only others to get a better understanding of the monitor’s raw performance but also allows the improvement of the algorithms with scientific partners.

A few important notes on how AirGradient implements these algorithms:

• All our current monitors with current firmware transmit data only as unaltered raw data to our server. The data is also saved as raw data in our database, which allows us to apply retroactive improvements in our correction algorithms.
• Data Export and the AirGradient API only provides raw data. Thus, the correction needs to be manually applied on that data.
• We are currently implementing the correction formula on the dashboard and monitor display. The status can be seen below.
• The temperature and relative humidity of the indoor monitor AirGradient ONE are very close to the reference and, therefore, will not be corrected. Only the outdoor monitor Open Air needs correction.
• PM₁ and PM₁₀ are currently not the focus, so no correction algorithms have been developed.
• CO₂ has an automatic baseline ongoing calibration; therefore, no correction algorithms are needed.
• VOC and NO_X are indexes based on relative changes of the pollutants and, therefore, do not need compensation algorithms.

Implementation Status

N/R: Not Required

Correction Formulas

Correction for PM_2.5

Extensive testing was carried out through our global co-location project. In the end, we deemed that the EPA correction formula developed for the Purple Air sensor (which contains the same PM module) performs very well.

We are currently writing a report on how the PM algorithms were tested and compared and will publish it here soon.

This correction formula splits up the algorithms into bands. Note that negative values could result from the formula. In this case, it is recommended to set them to 0.

AG_raw <30:

PM_2.5 = [0.524 x AG_raw] – [0.0862 x RH_raw] + 5.75

30≤ AG_raw <50:

PM_2.5 = [0.786 x (AG_raw/20 - 3/2) + 0.524 x (1 - (AG_raw/20 - 3/2))] x AG_raw – [0.0862 x RH_raw] + 5.75

50 ≤ AG_raw <210:

PM_2.5 = [0.786 x AG_raw] – [0.0862 x RH_raw] + 5.75

210 ≤ AG_raw <260:

PM_2.5 = [0.69 x (AG_raw/50 – 21/5) + 0.786 x (1 - (AG_raw/50 – 21/5))] x AG_raw – [0.0862 x RH_raw x (1 - (AG_raw/50 – 21/5))] + [2.966 x (AG_raw/50 –21/5)] + [5.75 x (1 - (AG_raw/50 – 21/5))] + [8.84 x (10^-4) x AG_raw²x (AG_raw/50 – 21/5)]

260 ≤ AG_raw:

PM_2.5 = 2.966 + [0.69 x AG_raw] + [8.84 x 10^-4 x AG_raw²]

Correction for Temperature and Relative Humidity

An extensive analysis with more than 200,000 temperature points was done, and the following correction algorithms were derived. More information about the development of these algorithms can be accessed here.

Only the AirGradient Open Air Outdoor Monitor needs correction. The indoor monitor AirGradient ONE has a different ventilation design and is therefore very accurate with the raw data.

Temperature

Below 10°C: T(calib) = T(raw)*1.327 - 6.738
Above 10°C: T(calib) = T(raw)*1.181 - 5.113

Relative Humidity

RH(calib) = RH_raw * 1.259 + 7.34

If a value is >100%, it will be set at 100%.