Monthly Archives: June 2018

PollutionPi: Project Proposal for an Indicative Air Quality Index Compatible Raspberry Pi Powered Air Quality Station

As a part of my volunteer position working on air pollution issues at CPI Fondacija located in Sarajevo, Bosnia and Herzegovina, I put together a project proposal for the design and prototyping of an inexpensive air quality station. This prototype would have provided the basis for building a network of air quality sensors that can provide indicative air pollution sensing in Sarajevo and other areas of Bosnia and Herzegovina. The aim was two fold. To advocate for action that curbs the severe air quality issues Bosnia and Herzegovina faces, and to provide awareness on how acute the air quality issue is in the region.

Unfortunately the project never went ahead, but I have decided to publish the project proposal here as I believe the project had merit and I did extensive research that I included in the proposal. I also think it is good to use in the sense portfolio as it shows some of my research and project design capabilities.

PollutionPi: An Indicative Air Quality Index Compatible Raspberry Pi Powered Air Quality Station

24 Hour Air Pollution Data in Federation of Bosnia and Herzegovina

In an attempt to make air pollution data in the Federation of Bosnia and Herzegovina more accessable to citizens, the Federal Hydrometeorological Institute of Bosnia and Herzegovina (FHMZBIH) has kindly given access to validated 24 hour air pollution data from the years of 2015, 2016, and 2017. The following interactive chart allows citizens to view this data.

24 hour data better shows the cycles of pollution across the Federation of Bosnia and Herzegovina, with a particular emphisis of poor air quality within the winter months.

Mobile Data

PMSA003 Low Cost Air Pollution Sensor Accuracy; An Attempt At Calibration in Sarajevo

In the last five or so years, affordable low cost air pollution sensors have become available on the market resulting in an explosion of open source online projects that measure the ambient quality of air. Much discussion and research has taken place within academic and government environmental organisations as to how accurate such sensors are and whether they have a place in the enforcing of environmental regulation, or even providing an indication of air quality trends to citizens.

As a part of project I am currently involved in with a NGO located in Sarajevo, I constructed multiple boxes that house low cost sensors which measure PM10 pollution. These boxes followed a design created by a locally based start up called CityOS. CityOS have resources available on their website including parts lists, a step-by-step guide to assembling an affordable air quality sensor that measures PM10, PM2.5, and PM1 particulate matters, a code base for the project, and an interface to view and download measured data. In order to gain a greater insight in to how accurate designs using these kinds of sensors are, I decided to attempt to get some comparison data on these sensors by using sensor collocation. The CityOS designed sensors will be placed in the immediate vicinity to professionally calibrated devices maintained by the Federal Hydrometeorological Institute of Bosnia and Herzegovina (FHMZBIH). With this data, I intend to attempt to calibrate these sensors as to increase their accuracy.

The PMSA003 Particulate Matter Sensor

The Plantower PA003

The Plantower PMSA003 is used in this particular CityOS sensor design. This sensor is available from a variety of online websites and generally costs between $15AUD and $20AUD a peice. No english datasheet exists for the device as far as I could find, however a datasheet written in Mandarin is available here.

The datasheet specifies a PM2.5 accuracy of +-10ug/ms when PM2.5 concentrations are below 100ug/m3, and a +-10% accuracy when PM2.5 concentrations are between 100ug/m3 and 500ug/m3. Already this is a little worrying, as according to EU and BiH regulations healthy PM2.5 concentrations exist below an annualised average of 25ug/m3, and healthy concentrations of PM10 exist bellow an hourly average of 50ug/m3. Having such a large inaccuracy at lower concentrations would likely result in unacceptable measuring accuracy for the majority of time. No PM10 accuracy information was included in the datasheet, which is problematic as this is the precise particulate matter that will be measured during the test.

The PMSA003 uses the method of laser diffraction to detect how much particulate matter is in the air. Laser diffraction typically requires calibration with a sensor using an alternative detection method such as Beta Attenuation due to the way different pollution diffracts light, and the highly sensitive nature of the technique to meteorological factors such as humidity.

The CityOS “Boxy” Sensor

The CityOS “Boxy” design includes a protective enclosure that houses various electronic components such as the ESP8266, as well as a DHT22 digital temperature and humidity sensor, and the PMSA003. The bottom of the enclosure is open so air can freely flow to the PMSA003, and so the electronics can be somewhat ventilated. The Boxy unit sends the data via a wifi connection, where the data can then be accessed on the CityOS or by utilising the CityOS API. Particulate matter measurements are taken every 60 seconds. When accessing the data via the CityOs API, the measurements are averaged to an hourly value.

The DHT22 sensor datasheet specifies a temperature reading accuracy of +-0.5 degrees Celsius, and a +-2% relative humidity accuracy.

The Reference PM10 Station and Test Setup

Verewa F-701 PM10 reference station

The Sarajevo based FHMZBIH kindly allowed us to place six CityOS Boxy units in the vicinity of their professionally calibrated Verewa F-701 PM10 measuring station so we could use it’s results as a calibration reference. This measuring station is located at the FHMZBIH offices in Bjelave, Sarajevo. The station completes a measurement once an hour and the result is accessible to the public on the FHMZBIH website.

The housing for the CityOS Boxy sensors.

The six units were placed approximately four metres away from the reference station within a basket in a small housing that contained vents that allowed air flow and sheltered from the elements such as rain.

Six Boxy units inside basket.

In addition to reference PM10 data,  reference temperature and humidity data were also to be recorded in order to correlate the CityOS Boxy PM10 measurement accuracy with various weather conditions. The accuracy of the DHT22 temperature and humidity measurements were also to be compared to this data. The temperature and humidity data could potentially be used to calibrate the PMSA003 sensor regardless of the weather conditions.

The Results

Mobile Data

The above interactive chart shows data collected over a ten day period from sources including data sourced from six CityOS Box units, FHMBIH PM10 reference station, and FHMBIH temperature/humidity reference station.

Analysis

The PM10 measurement accuracy of the PMSA003 particulate matter sensor housed in the CityOS Boxy enclosure was poor in almost all conditions presented while the testing was conducted.  By navigating the above interactive chart, the following conclusions can be made:

  •  Over the ten day test period the average error of the PMSA003 sensor in a CityOS Boxy enclosure was approximately +- 60% when compared to the reference PM10 measurements.
  • PM10 measurement error is significantly higher at lower PM10 concentrations, thus confirming the datasheet specification of significant errors at particulate matter concentrations below 100ug/m3.
  • There is a trend of higher PM10 measurement error as humidity increases, however there is no significant statistical correlation that can be utilised for calibration.
  • There is no significant statistical correlation between PM10 measurement error and temperature.
  • PMSA003 and DHT22 measurements were broadly uniform across all units.
  • The DHT22 temperature measurements had an average accuracy of approximately +-15%. The discrepancy of this accuracy and that reported on the DHT22 datasheet is likely to be due to the sensor placement in the CityOS boxy enclosure.
  • The DHT22 humidity measurements had an average accuracy of approximately -+50%. The discrepancy of this accuracy and that reported on the DHT22 datasheet is likely to be due to the sensor placement in the CityOS boxy enclosure.

From these conclusions, it is apparent that calibration for measurements under 100ug/mg may not be possible. While there are some correlations between PM10 measurement accuracy and PM10 concentration and humidity, the correlations are not represented in a uniform in a way that may be useful for the calibration of the sensors.

Conclusion

Due to the high measurement errors at lower particulate matter concentrations that do not correspond to bias errors, it is not possible to calibrate PMSA003 sensor to provide more accurate data at concentrations below 100ug/m3.  This calls in to question the usefulness of the PMSA003 sensor in providing accurate particulate matter measurements.

However, it may still be possible to use the PMSA003 in a useful way to inform individuals about dangerous levels of particulate matter concentrations in the air when hourly averaged values indicate concentrations above 30ug/m3. The United States Environmental Protection Agency (EPA), in partnership with the Village Green Project, conducted research in to the usability of low cost sensor in informing the public of poor air quality within communities. By analyzing the data measured by low cost sensors they came up with a scaling system that provides useful information to the public for PM2.5 measurements, while taking in to account the inaccuracies of low cost sensors.

The Village Green Project PM2.5 scaling system.

For PMSA003 measurements to be useful a similar scaling system should be utilised, and individual measurements should be treated with skepticism. The data analysed from this test seems to corroborate with this scaling system in terms of the accuracy of the measurements at different concentration levels. More qualitative analysis of the test data is required in order to make an similar scaling system catered towards the  PMSA003, however at first analysis the Village Green Project scaling system appears to be a good solution to the issue of poor measurement accuracy.