**1. Introduction**

#### *1.1. Summary of Problem*

High concentrations of air pollution are due to fine solids, gases, or liquid aerosols locally releasing into the atmosphere or being produced at a faster rate than the environment can dilute, absorb, or dissipate the material [1]. If the rate of production is su fficiently high, substances can build up and reach a high concentration in the air that can contribute to a host of adverse health e ffects for humans

such as cardiovascular mortality and respiratory distress [2–6]. Miller et al. found an increase of 10 μg/m<sup>3</sup> in particulate matter mass concentration was associated with a 14% (95% confidence interval, 3%–26%) increase in nonfatal cardiovascular events and with a 32% (95% CI, 1%–73%) increase in fatal cardiovascular events [7]. Exposure to air pollution whether long or short can lead to asthma, decreased lung function, and infections of the respiratory system. Goss et al. found PM2.5 was associated with statistically significant declines in lung function and an increase in the odds of two or more pulmonary exacerbations in patients >6 years of age with cystic fibrosis [8]. Woodru ff et al. found that each 10 μg/m<sup>3</sup> increase in PM2.5 was associated with a near doubling of the risk of post neonatal death because of respiratory causes [9]. Hoek et al. found the risk of cardiopulmonary mortality nearly doubled for individuals who lived within 100 m of a freeway or within 50 m of a major urban road [10]. Clearly, particulate matter presents a clear and present danger to human health.

Aerosols are found everywhere including in the air over oceans, deserts, mountains, forests and ice, with the most abundant being from natural sources such as sea-spray and wind-blown soil dust [11]. For instance, dry areas of North Africa emit 800 Tgy−<sup>1</sup> of soil dust each year and summertime winds can transport the Saharan dust across the Atlantic Ocean, the Caribbean, and to southern North America [12–15]. Urban areas add to the airborne particulate mass through direct emissions into the atmosphere and through emitting gases that can react in the atmosphere to form secondary aerosol. A significant body of research has gone into understanding and controlling the chemical and physical transformations that lead to secondary aerosol formation [16–18]. Despite the grea<sup>t</sup> success of air quality improvement programs implemented in Western cities over the preceding 50 years, urban environments typically have higher levels of particulate pollution compared to rural locations as cities remain global hubs of fossil fuel combustion and direct emissions [19–23]. An emerging threat is found in some of the world's poorest cities as the World Health Organization (W.H.O.) has recently reported 98% of cities in low and middle income countries with more than 100,000 inhabitants do not meet WHO air quality guidelines [24]. Aerosols are also present indoors and homemakers in economically disadvantaged environments can often be exposed to very high levels of particulate matter when using home cook-stoves to prepare meals [25,26]. Monitoring and controlling human exposure to PM2.5 remains a crucial scientific challenge and this focus requires the continued development and application of portable and low–cost sensing platforms for widespread application.

#### *1.2. Current Work in the Field*

Historically, PM2.5 concentrations have been measured using either inertial impactors, filter based sampling, or sophisticated laboratory-based measurement devices [27–31]. However, such approaches are either too expensive, too labor intense, too slow, too limited in scope, or all of the above to address the needs of the problem. A new paradigm is needed in which many end users can access accurate, real-time data at very low cost to society [32,33].

Several brands (Shinyei, Sharp, Nova, Plantower, Wuhan, Alphasense, Air Beam, etc.) of low-cost electronic sensors that use light scattering can be used to monitor particulate matter disbursed in air. These sensors come at di fferent price points and sizes but are designed to use low power and can operate at high measurement frequency. The common limitation in the sensors is the accuracy and the reproducibility of the data, and very limited ability to measure particles below 200–300 nm in diameter due to low scattering cross section. In addition, low-cost sensors often have limited linear dynamic range, and often plateau or 'max out' at a few hundred microgram per cubic meter concentration. This feature of devices may underestimate the impact of high exposure events.

Currently, one of the most successful integrated approaches to particulate matter monitoring is called Airbeam (habitatmap.org). In this project, a palm-sized device costing \$250 measures particulate pollution that a user is exposed to by a light scattering method and transmits data to a co-located Android device through a Bluetooth connection. The wireless connection on the secondary device and an Android OS App is used to 'AirCast' collected data to a central data server and maps are created by combining various end users' data streams. The integrated approach demonstrates the potential of the crowd-sourced architecture.

The excitement generated by the initial emergence of low-cost sensing platforms such as 'Airbeam' has led the Environmental Protection agency (EPA) of a major North American country to begin evaluating the performance of low-cost sensors co-located with sensors their governmen<sup>t</sup> considers "reference" and "equivalent" methods (see https://www.epa.gov/air-sensor-toolbox). In an exclusive project, in 2019 this EPA began a study aimed to tackle questions about long-term performance of low-cost air sensors, an area that is rarely (if ever) explored in academic settings. The exclusive group of EPA funded sta ff are currently evaluating six di fferent models of low-cost air sensors, placed at seven diverse locations throughout states within North America. The locations have diverse climates and air quality conditions, assuring that the project produces a dataset that investigators may use to assess how weather conditions a ffect sensor signals and long-term performance.

In Gunawan et al., a suite of portable sensors (PM2.5, PM10, CO, O3) were used in Malaysia to derive the local air pollution index (API) at an end user's location [34]. The API was reported to the user in nearly real-time thru an LCD display. For carbon monoxide a model MQ-9 sensor was employed, while ozone was monitored with the MQ-131 gas sensor. The Sharp dust sensor was used to measure PM10 while a Shinyei PPD42NS was used to derive PM1-2.5. The set-up was controlled by the Arduino microcontroller, and the authors reported a cellular modem and Global Positioning System (GPS) card could be added if desired. However, wireless communication was not a focus of the work. Unfortunately, the authors were unable to perform proper calibration of each sensor and had to rely on suggested calibration equations from the manufacturer. In addition, laboratory validation measurements were not completed and data quality unassessed.

In Reilly et al., an a ffordable monitoring device costing ≈\$280 was constructed [35]. The device can send and collect data to a real-time mapping program wirelessly using the Global System for Mobile Communications (GSM) to a server, or as a text message to people nearby. The device consisted of the Redboard Arduino clone, a GSM board, a carbon monoxide sensor, an ozone sensor, and Sharp dust sensor for particulate matter (PM), a fan to promote airflow, and instrument case. While the authors report successful implementation of their device, environmental sensors were not calibrated nor rigorously tested in a laboratory. Instead, the authors used the recommended calibration equations from the product manufacturers to convert electrical signal to environmental measurement. Suspiciously high and constant PM levels were reported (approx. 150 μg/m3) and ozone levels did not exhibit the characteristic diurnal pattern that is observed nearly everywhere in the troposphere. Consequently, despite the approach being successfully implemented, the quality of environmental sensor results is questionable.

In Zamora et al., Plantower AMS A003 sensors were tested in the laboratory to measure particulate matter of various composition in a series of well-defined and controlled experiments [36]. The sensors were precise; however, when the sensor results were compared with a reference method, accuracy varied as a function of aerosol composition and humidity. The sensors were highly precise with a R<sup>2</sup> values greater than 0.86 for all sources. However, the accuracy had a large range from 13% to greater than 90% compared with reference instruments, depending upon aerosol type. The sensors were more accurate when the particulate matter was spherical and smaller than 1 μm in size. The sensors' accuracy was greatly a ffected by the humidity. The work of Zamora et al. contributes significantly to the field because it draws attention to the limits of single angle nephelometry to estimate PM mass concentration vis-à-vis the aerosol composition and shape. While the sensors were able to provide usable data when in motion or in high or low temperatures, the focus of this manuscript was refreshingly on data quality. Results sugges<sup>t</sup> that portable sensors should be calibrated versus an accepted reference method using authentic aerosol as a sample.

Liu et al. takes this approach by examining performance of Nova particulate matter sensors (SDS011) [37]. The sensors were tested on ambient samples in Oslo, using a co-located, o fficial air quality monitor as a reference method. All of the sensors had results that were similar including inter-sensory correlations with R<sup>2</sup> values greater than 0.97. Again, high humidity greatly a ffected the sensors. There was a linear relationship between the sensors and the reference monitor. However, the R<sup>2</sup> varied over the range of 0.55 to 0.71. When a data correction using relative humidity and temperature was used, the R<sup>2</sup> value for each sensor increased from 0.71 to 0.80, 0.68 to 0.79 and 0.55 to 0.76, respectively. These sensors were also limited by the environmental conditions and can have high error if used outside of the manufacturer specifications. This work draws attention to the need to calibrate versus an accepted gravimetric reference method with authentic aerosol and the need to correct measurements (or at least flag them) for conditions of high humidity.

Liu et al. have compared Shinyei, Sharp, and Oneair optical sensors in laboratory experiments with polydisperse particles of a variety of compositions, concentration, and mean size [38]. These authors found the mass concentration normalized response of all optical sensors clearly changed with mean particle size even when considering the narrow range of very small particles between 70–95 nm. The composition of particles was even more crucial—particles of methylene blue, sodium fluorescein, and sodium chloride all produced sensitivities di ffering by at least 50% (with NaCl being lowest). These results clearly demonstrate that choice of calibrant aerosol is crucial for obtaining accurate sensor mass calibrations. The current manuscript is not exempt from such considerations or potential errors.

#### *1.3. Contribution of the Current Work*

The current study aims to build upon previous work from our laboratory in which a field-portable device for logging PM2.5 mass concentration data was developed [39]. The previous device also used the Arduino, and a Sharp sensor with nephelometric detection, but logged all collected data to an SD card. This prevented real-time feedback to the user and complicates integrating/multiplexing many sensors together to form a network of streaming data. This limitation has recently been addressed thru our laboratory's development of an open-source data acquisition platform we have called *logIT* that wirelessly transmits data over the 3G cellular network [40]. Source code to implement *logIT* is available online for the user community [40]. The *logIT* platform has allowed multiplexing multiple sensors (here n = 3) to demonstrate multi-user simultaneous measurements of PM exposure throughout a mid-sized city. In addition, we build upon the previous literature work using the Sharp GP2Y1010AU dust sensors [34,35] by performing laboratory measurements to characterize and constrain device performance. This was not reported in [34,35] and results presented within this manuscript will allow the community to make more informed decisions regarding the performance characteristics of Sharp GP2Y1010AU dust sensors. The current manuscript also serves as a companion article to [38], as the current work highlights signal to noise considerations and intercomparisons of individual Sharp dust sensors themselves, while Liu et al. [38] provide much more through insight into calibration of low-cost sensors against a gravimetric equivalent method and the crucial e ffect of calibrant composition and size.
