Low-cost air quality monitoring networks can potentially increase the availability of high-resolution monitoring to inform analytic and evidence-informed approaches to better manage air quality. This is particularly relevant in low and middle-income settings where access to traditional reference-grade monitoring networks remains a challenge. However, low-cost air quality sensors are impacted by ambient conditions which could lead to over-or underestimation of pollution concentrations and thus require field calibration to improve their accuracy and reliability. In this paper, we demonstrate the feasibility of using machine learning methods for large-scale calibration of AirQo sensors, lowcost PM sensors custom-designed for and deployed in Sub-Saharan urban settings. The performance of various machine learning methods is assessed by comparing model corrected PM using k-nearest neighbours, support vector regression, multivariate linear regression, ridge regression, lasso regression, elastic net regression, XGBoost, multilayer perceptron, random forest and gradient boosting with collocated reference PM concentrations from a Beta Attenuation Monitor (BAM). To this end, random forest and lasso regression models were superior for PM 2.5 and PM 10 calibration, respectively. Employing the random forest model decreased RMSE of raw data from 18.6 μg/m 3 to 7.2 μg/m 3 with an average BAM PM 2.5 concentration of 37.8 μg/m 3 while the lasso regression model decreased RMSE from 13.4 μg/m 3 to 7.9 μg/m 3 with an average BAM PM 10 concentration of 51.1 μg/m 3 . We validate our models through cross-unit and cross-site validation, allowing analysis of AirQo devices' consistency. The resulting calibration models were deployed to the entire large-scale air quality monitoring network consisting of over 120 AirQo devices, which demonstrates the use of machine learning systems to address practical challenges in a developing world setting.

Air pollution is considered a major public health risk globally, and the global South including sub-Saharan Africa face particular health risks, but there is limited data to quantify the level of pollution for different air quality contexts. The COVID-19 lockdown measures led to reduced human activities, and provided a unique opportunity to explore the impacts of reduced activities on urban air quality. This paper utilises calibrated data from a low-cost sensor network to explore insights from the diverse ambient air quality profile for four urban locations in Greater Kampala, Uganda before and during lockdown from March 31 to May 5 2020, highlighting the uniqueness of air pollution profiles in a sub-Saharan African setting. All locations saw year to year improvements in 24-hour mean PM2.5 between 9 μg/m3 and 25 μg/m3 and correlated well with reduction in traffic (up to approx. 80%) and commercial activities. The greatest improvement was observed in locations close to major transport routes in densely populated residential areas between 8:00 pm and 5:00 am. This suggests that the reduction in localised pollution sources such as nocturnal polluting activities including traffic and outdoor street cooking characteristic of fast-growing cities in developing countries, coupled with meteorological effects led to amplified reductions that continued well into the night, although meteorological effects are more generalised. Blanket policy initiatives targeting peak pollution hours could be adopted across all locations, while transport sector regulation could be very effective for pollution management. Likewise, because of the clustered and diffuse nature of pollution, community driven initiatives could be feasible for long-term mitigation.

No abstract

<p>Increasing awareness of air pollution requires access to timely and reliable air quality data and information, and yet many African cities lack effective air quality monitoring infrastructure, largely because of the resource constraints of establishing and managing a continuous monitoring network. Low-cost sensor platforms have the potential to close the air quality data gaps in resource-strained settings such as Africa, but the continued lack of accessible and reliable infrastructure for data management is a major hindrance to effective air quality management.</p><p>Moreso, managing a large Internet of Things (IoT)-based sensor network can be complex, and the demand for a case-specific and highly customizable platform, coupled with its conceptualization & implementation complexities, renders most existing IoT platforms ineffective. There is a need for a platform infrastructure for continuous support and management of air quality data with a high spatial and temporal resolution to facilitate sophisticated analysis; while taking care of the associated structural challenges of low-cost sensors.  The AirQo platform, a robust could-native software is a novel communityaware digital platform for managing large-scale air quality networks, applicable in resource-strained environments. This customisable and scalable platform attempts to address the data access challenges, with capabilities to become a ‘one-stop centre’ for management of other third party IoT sensor networks. Different interfaces through mobile application, web-based dashbord and platform cater for diverse data needs. The robust approach enables decision makers and other stakeholder communities have access to timely and quality assured air quality data. Using a set of metrics, user-experiece can be computed and compared with existing IoT management platforms. Software design considerations including (1) Multi-tenancy, (2) Data pipeline, (2) Sharded Database Cluster, (3) Microservices architecture, (4) Containerized deployment, and (5) Interoperability are recommended to support replication in other use-cases.</p><p> </p><p> </p>