Democratizing Air Quality Monitoring in Tehran

Team: CEO: Arian Bakhtiarnia, Garage Lead and Product Manager: Amirreza Azadeh, Electronics Engineer: Abolfazl Ghamgosar, Design Engineer: Abbas Sohrabi, Software Engineer: Farzad Karamouz, Data Analyst: Melika Farahani

Air pollution is among the most critical health problems created by the uncontrolled industrialization of large cities. According to The World Bank, air pollution is the fourth-leading risk factor for deaths worldwide, and according to the Institute for Health Metrics and Evaluation at the University of Washington, it is the 9th health risk factor that drives death and disability combined for Iranians.

Among 26 megacities of the world, Tehran, the capital of Iran, which usually has more than 100 unhealthy days during a year, ranks 12th most polluted megacity with four times more pollution than the recommended tolerable amount.

Tehran’s municipality reports every year around 5000 youths die because of air pollution. The economic damage by Tehran’s air quality to the health system of Iran is estimated to be $2.6 billion without considering its financial damage through decreasing agricultural products, decreasing quality of life, days that schools, government and all offices are closed, and decrease in cognitive function due to high pollution.

Due to the issue’s significance, at Balad, the market-leader consumer maps and navigation app in Iran, Balad Garage was tasked to investigate how we can positively contribute to raising awareness about the problem. Our research showed that Balad’s role could include these parts:

1. How can we increase citizen accessibility of the already existing data?
2. How can we increase the density of data to better understand sources and patterns of pollution?
3. How can we engage citizens in participatory and citizen science practices of measurement?

Increasing accessibility of the city’s data for millions of users

In response to the first challenge, we decided to present the existing air quality data to millions of users of Balad mobile app. The sources included official stations of the municipality and the Department of the Environment which both maintain and manage different stations across the city. Despite open API claims of the municipality on their website, the negotiations between the company and these organizations took a year to receive an API from the city and nothing from the Department of Environment. Lesson learned: multi-stakeholder public-private collaboration is not as predicted in every country. Although, a few months after releasing the air quality feature in Balad app with the city’s API, we used the data presented on the DoE website without an official API. Interestingly, there was no opposition. Users started to use the feature in an unprecedented number. And we could see how in the more polluted dates, the usage jumps and the discussion of the measurements circulates in the society and social media.

The expected outcomes for the implementation of low-cost sensors

In response to the second challenge, increasing the density of data to understand sources and patterns of pollution better, we decided to evaluate the potential of low-cost sensors as an addition to the network of the city’s official air quality monitoring stations’ data. With such sensors, we hoped to:

1. Achieve hyper-local measurement instead of regional measurement to gain a spatial resolution.
2. Increase the number of stations (low-cost) and eventually cover marginalized areas despite financial limitations.

We were excited by the potential of the kit in responding such questions:

“How air pollution is different…”:

1. inside and outside Balad office and the city’s station?
2. at various highways and tunnels?
3. at parks and green areas compared to the surrounding neighborhood?
4. across transportation modes, inside and outside buses, inside subway stations, inside subway, walking on street levels, inside and outside of a driving car?
5. in proximity to main highways and away from them in the same neighborhood?
6. at certain locations during the day and night and weekdays versus weekdays?
7. and, Where are the best spots in the city for physical exercise during the healthy days?
8. and, How kindergartens, schools, and hospitals are or are not exposed to unhealthy pollution?

In response to the third challenge, engaging citizens in participatory and citizen science practices of measurement, we aimed to create a moving network of kits, developed by students and hosted by them to send data during the kit’s life-span. We discussed the idea with university students and already have volunteers to set up the kit on their backpacks. Through this strategy, in addition to engaging citizens in the project, we can also:

1. Repeat the measurements in specific paths with a low number of kits, since students navigate the same route to the university and workplace every day.
2. Measure and broadcast wherever is important for the volunteers, in contrast to top-down decision-making of the routes and locations of the stations.

With the mobile app feature already in the hands of many users, we could easily add the mentioned layers of information to the city’s stations data layer. The moving networks’ data could be visualized as line segments of data. The next stage of the project was viability evaluation of running the project.

Therefore, we decided to design, develop, and test a portable air quality monitoring kit to measure air pollutant density. We used Alphasense electrochemical sensors to measure CO, SO2, NO, NO2, and O3, and Alphasense Optical Particle Counter sensors to measure PM1, PM2.5, and PM10.

Balad Kits’ electronics hardware included:

1. BeagleBone Black
2. Custom Design PCB (Control Board)
3. Alphasense 2-Way AFE A4 sensor for CO & SO2
4. Alphasense 3-Way AFE A4 sensor for NO2 & O3 & NO
5. Alphasense OPC-N3 (Optical Particle Counter) for PM1, PM2.5, and PM10
6. Bosch BME280 sensor for Temperature, Humidity, Pressure
7. 4G Module
8. Ublox EVK-M8U GNSS Positioning Kit
9. Xiaomi Redmi 10000 mAh Power Bank
10. Fan

6. Calibration

Field calibration of electrochemical sensors and optical particle counters in proximity to the municipality’s station

Due to the low-cost nature of electrochemical sensors and particle counters, they need to be calibrated with a reference, either clean air, or a known air quality. Among various methods, we chose to calibrate our two kits of sensors by placing them next to second region of the city’s stations for three months to gather data with different temperatures and humidity levels.

We expected to see similar outputs for the kits. The electrochemical sensors for measuring the gases had almost similar results despite some offsets but the second particle counter had extreme noises. After communicating with Alphasense we realized that it needs to be replaced and we continued the calibration only with one of the OPCs.

For calibration of each pollutant measurement, we tested various algorithms and parameters.

Data readings: voltage (for electrochemical sensors), ppb (for particle counters) per second, calculation: hourly average, ppm (reference) / Period: 11 days / Algorithms: Naive Regression Models, ARIMA-X, Linear Regression, KNN Classifier, Trained with 80% of data and predicted 20% / Features: Temperature, Humidity, raw WE & raw AE sensor outputs, corrected WE & AE based on Alphasense formulas.

We then compared errors of each method relative to the municipality reference. Linear Regression worked best for NO2, O3, CO, PM10 with temperature, humidity, and WE & AE as their features.

[RMSE = The Root Mean Squared Error] // NO2 RMSE = 6.20, error = 10% // O3 RMSE = 8.71, error = 30% // CO RMSE=0.68, error = 35% // PM10 SME = 638.52, error = 26%

Overall, some of our results had more accuracy than similar works and research papers (1, 2, 3) and some need more work. We suggest gathering more data during other weather conditions and seasons.

7. Conclusion

Evaluating the potential of long-term low-cost/high-density measurement of air quality in Tehran

In many parts of the world, there are success stories and hype around the potential of low-cost/high-density measurement of urban air quality. However, after a period of research, development, and experimentation, we conclude that using such technology with the intention of implementing a high quantity network of kits for a long-term use has not been not viable, specifically after the country’s recent economic collapse, we faced unforseen political and economical events in the country that changed the situation of the project.

Firstly, Alphasense sensors have been purchased with complexity due to the international sanctions against Iran, and due to their short life span (36 months), it is a matter of question whether any company and non-governmental institute in Iran can periodically purchase such high-quality parts.

Secondly, although these sensors are low-cost compared to the city’s industrial stations, in reality, they are not anymore considered as affordable as they are in more developed countries. There might be a chance in the future with technology advancements in providing such reliable low-cost sensors with lower prices.

Thirdly, the sensors’ need for monthly visits and occasional calibration, requires an operation team and human resources not predicted by the team and not discussed much in similar projects. From this point of view, we do not find them very low-maintenance.

Lastly, we still do find such low-cost sensors useful to be used for two intentions in the country:

1. Small scale participatory and citizen-science projects in Iran are achievable as we experienced an unimaginable excitement of university students’ feedback in working with sensors and exchanging experience.

2. It is viable to run short-term research projects with defined questions but not with the intention of long-term monitoring. Therefore, the need for a long-term map-based product in our company required the second and third challenges of the air quality project to be discontinued, while our users are still using the provided mobile app feature.

1. Smart Citizen Team. Index – iSCAPE Sensors Documentation. [cited 12 Mar 2020]. Available: https://docs.iscape.smartcitizen.me/Sensor%20Analysis%20Framework/
2. Maklin C. ARIMA Model Python Example — Time Series Forecasting. In: Medium [Internet]. Towards Data Science; 25 May 2019 [cited 14 Mar 2020]. Available: https://towardsdatascience.com/machine-learning-part-19-time-series-and-autoregressive-integrated-moving-average-model-arima-c1005347b0d7
3. Anonymous, Anonymous. Use of electrochemical sensors for measurement of air pollution: correcting interference response and validating measurements. doi:10.5194/amt-2017-138-rc1