Clearing the Air with AI: Revolutionizing Air Quality Monitoring Through Machine Learning

Introduction

The previous posts on this blog have focused on explicating the harmful effects of air pollution in various forms and on various beings. Nonetheless, innovative technologies are emerging that seek to mitigate the harmful impacts of air pollution. One interesting area that is rapidly developing is the use of artificial intelligence (AI), machine learning (ML), and deep learning (DL) to aid in air quality monitoring and forecasting. Such technologies can better inform pollution governance through providing more robust information regarded pollution trends ahead of time.

The development of the Internet of things (IoT) has enabled a deployment of sensor networks that provide real-time data about air pollution levels in urban environments [1]. However, the raw data obtained from these sensor networks must be analysed to provide insight regarding present and future fluctuations in air quality. That’s where AI and machine learning come in. The data collated from sensor networks can be used to train models that predict air pollution levels in real-time, and identify patterns and trends that would be difficult to discern manually [2]. Crucially, such comprehensive training of these models also requires additional data beyond mere air quality, such as meteorological data, temporal data, spatial data, built environment and population variables, satellite-retrieved data, weather forecast data, and chemical component forecast data [3].  Moreover, Kumar et al (2020) found that using machine learning algorithms with low-cost air quality sensors enables accurate and cost-effective air quality monitoring [4]. Further studies have shown that machine learning models can improve air quality forecasting accuracy when compared to traditional methods [5] .Therefore, this allows decision-makers to respond more effectively to changing air quality conditions, whether it’s by adjusting traffic flow, regulating industrial processes, or alerting residents to potential health risks.

Moreover, researchers compared the accuracy of the deep learning model- long short-term memory (LSTM)- with traditional statistical modelling techniques for predicting PM2.5 concentration levels [6]. Accordingly, they illustrated how the predicted levels of PM2.5 were closer to the true value of PM2.5 when using the LSTM model as compared to traditional models, but also cautioned that such accuracy requires robust data regarding the air quality of the region being investigated. In fact, a hybrid of multiple ML models enables even more accurate predictions (Figure 1) since they may complement the predictive ability of one another. For instance, during prediction of PM2.5 concentration, CNN (Deep Convolutional Network) provides more spatial information, while LSTM provides temporal information [7].

Figure 1 Comparison between PM2.5 forecast based on LSTM-CCN hybrid model and actual historical PM2.5 concentration for 10 days (Bekkar et al, 2021)

 

Further Practical Implementations

The combination of historical data from air quality sensors and public health information from epidemiological reports can enable the identification of high-risk localities/populations facing air pollution. Consequently, scientists are advocating for the development of a big data ecosystem that integrates data from various sources to optimise forecasting techniques for localised contexts [8]. Moreover, such technologies can also be refined for further personalisation at the level of individual consumers. Schürholz et al (2020) incorporated user specific conditions such as sensitivity to particular pollutants or respiratory conditions such as asthma, as well as situational conditions such as traffic volumes and fire incidents within the training algorithm for their model [9]. This inclusion of context-aware computing (Figure 2) enabled a personalisation of the Air Quality Index (AQI) that was obtained, which they termed the “MyAQI”.

Figure 2 Context-aware prediction model proposed for MyAQI (Schürholz et al, 2020)

 

Moreover, the deployment of such technologies in urban contexts can potentially enhance public health outcomes by providing individuals real-time information tailored to their health needs regarding the air quality of their immediate surroundings, thereby enabling greater sensitivity while navigating different spaces. For instance, an individual suffering from asthma may choose to avoid a congested road with high levels of nitrogen dioxide and ozone based on information available on MyAQI. It also provides opportunity for citizens to hold governments and regulatory bodies accountable should air quality fall below the stipulated environmental quality standards. Finally, models may also be incorporated into the cloud server that assess the toxicity of real-time air quality and subsequently send out alerts if hazardous pollutant levels are present in the air (Figure 3)  [10].

Figure 3 Internet of Things based pollution monitoring network with an alert system (Asha et al, 2022)
Limitations

As air pollution worsens in our rapidly urbanising world, it remains essential to update our technology to obtain and assess more accurate information regarding pollution levels. Nonetheless, the AI and ML approach comes with its own set of limitations. For instance, the ability to provide personalised assessments using individual health profiles and location information may raise concerns of data privacy. Importantly, the accuracy of predictions made by AI and ML models depends significantly on the quality of the data used to train these models as well as the quality of data collected from the sensors. Ultimately, although comprehensive information obtained from these models can better inform policy decisions and regulatory frameworks, they must be integrated into larger schemes primarily targeted at reducing emissions at the source.

 

References

[1] Gulia, S., Shiva Nagendra, S. M., Khare, M., & Khanna, I. (2015). Urban air quality management-A review. Atmospheric Pollution Research, 6(2), 286–304. https://doi.org/10.5094/apr.2015.033

[2] Petry, L., Meiers, T., Reuschenberg, D., Mirzavand Borujeni, S., Arndt, J., Odenthal, L., Erbertseder, T., Taubenböck, H., Müller, I., Kalusche, E., Weber, B., Käflein, J., Mayer, C., Meinel, G., Gengenbach, C., & Herold, H. (2021). DESIGN AND RESULTS OF AN AI-BASED FORECASTING OF AIR POLLUTANTS FOR SMART CITIES. ISPRS Annals of the Photogrammetry, Remote Sensing and Spatial Information Sciences, VIII-4/W1-2021, 89–96. https://doi.org/10.5194/isprs-annals-viii-4-w1-2021-89-2021

[3] Iskandaryan, D., Ramos, F., & Trilles, S. (2022). Application of deep learning and machine learning in air quality modeling. Current Trends and Advances in Computer-Aided Intelligent Environmental Data Engineering, 11–23. https://doi.org/10.1016/b978-0-323-85597-6.00018-5

[4]  Kumar, K., & Pande, B. P. (2022). Air pollution prediction with machine learning: a case study of Indian cities. International Journal of Environmental Science and Technology. https://doi.org/10.1007/s13762-022-04241-5

[5] Zhang, B., Zou, G., Qin, D., Ni, Q., Mao, H., & Li, M. (2022). RCL-Learning: ResNet and convolutional long short-term memory-based spatiotemporal air pollutant concentration prediction model. Expert Systems with Applications, 207, 118017. https://doi.org/10.1016/j.eswa.2022.118017

[6] Gladkova, E., & Saychenko, L. (2022). Applying machine learning techniques in air quality prediction. Transportation Research Procedia, 63, 1999–2006. https://doi.org/10.1016/j.trpro.2022.06.222

[7]  Bekkar, A., Hssina, B., Douzi, S., & Douzi, K. (2021). Air-pollution prediction in smart city, deep learning approach. Journal of Big Data, 8(1). https://doi.org/10.1186/s40537-021-00548-1

[8] Shih, D.-H., To, T. H., Nguyen, L. S. P., Wu, T.-W., & You, W.-T. (2021). Design of a Spark Big Data Framework for PM2.5 Air Pollution Forecasting. International Journal of Environmental Research and Public Health, 18(13), 7087. https://doi.org/10.3390/ijerph18137087

[9] Schürholz, D., Kubler, S., & Zaslavsky, A. (2020). Artificial intelligence-enabled context-aware air quality prediction for smart cities. Journal of Cleaner Production, 271, 121941. https://doi.org/10.1016/j.jclepro.2020.121941

[10] Asha, P., Natrayan, L., Geetha, B. T., Beulah, J. R., Sumathy, R., Varalakshmi, G., & Neelakandan, S. (2022). IoT enabled environmental toxicology for air pollution monitoring using AI techniques. Environmental Research, 205, 112574. https://doi.org/10.1016/j.envres.2021.112574

Leave a Reply

Your email address will not be published. Required fields are marked *

Viewing Message: 1 of 1.
Warning

Blog.nus accounts will move to SSO login, tentatively before the start of AY24/25 Sem 2. Once implemented, only current NUS staff and students will be able to log in to Blog.nus. Public blogs remain readable to non-logged in users. (More information.)

Skip to toolbar